Evaluation of Tsunami Risk from Regional Earthquakes at Pisco, Peru

We evaluate tsunami risk for the port city of Pisco, Peru, where major liquefied natural gas facilities are planned. We use a compilation of instrumental and historical seismicity data to quantify the sources of six earthquakes that generated tsunamis resulting in minor inundation (1974) to catastrophic destruction (1687, 1746, 1868) in Pisco. For each of these case scenarios, the seismic models are validated through hydrodynamic simulations using the MOST code, which compute both flow depth on virtual offshore gauges located in Pisco harbor and the distribution of runup in the port and along the nearby beach. Space-time histories of major earthquakes along central and southern Peru are used to estimate recurrence times of tsunamigenic earthquakes. We conclude that Pisco can expect a metric tsunami, capable of inflicting substantial damage every 53 years, and a dekametric tsunami resulting in catastrophic destruction of infrastructures every 140 years. The last such event occurred 138 years ago. An important result of our study is that total destruction of the city of Pisco during the famous 1868 Arica tsunami requires an earthquake rupture straddling the Nazca Ridge, which thus constitutes at best an imperfect “barrier” for the propagation of rupture during megathrust events. This gives a truly gigantic size to the 1868 Arica earthquake, with a probable seismic moment reaching 10 dyne cm. Introduction and Background We present a pilot study quantifying tsunami risk along a portion of South American coastline including the port of Pisco, Peru (Fig. 1), where major liquefied natural gas (LNG) facilities are planned. The coastline of South America, located at the boundary between the Nazca and South American plates, features exceptionally large earthquakes, which have characteristically triggered major tsunamis inflicting severe destruction in both the near and far fields. This warrants critical assessment of tsunami risk in the context of the development of facilities such as the proposed LNG terminal at Playa Loberia. The city of Pisco, Peru, is located at 13.7 S, 76.2 W, 200 km southeast of Lima along the Peruvian shore, at the northern termination of a bay limited to the south by the Paracas peninsula (Fig. 1). The peninsula separates two segments of coastline with slightly different azimuths, and different histories of seismic activity: to the north, the central shore extends from Pisco to Chimbote (9 S) at an azimuth of N330 E, with seismic rupture recently expressed through moderately large earthquakes; to the south, the Peruvian southern shore, trending N305 E to the Arica Bight at 19 S, was the site of gigantic historical earthquakes, such as the famous 1868 event. A further important tectonic feature is the aseismic Nazca Ridge, in general, interpreted as a fossil hotspot track (Pilger and Handschumacher, 1981), which subducts along a 175-km segment of coastline, between latitudes 14.5 S and 15.5 S. We address the question of the tsunami risk at Pisco by examining the historical record of tsunami damage along the coast of central and southern Peru, from 9 S (Chimbote) to 19 S (Arica; now in Chile), building seismic models of the principal events involved and conducting numerical simulations of the run-up at Pisco (and other coastal locations) that these models predict. We then evaluate possible return periods for the main seismic events under consideration. These individual simulations make our approach significantly different from the recent work of Kulikov et al. (2005), based on the empirical concept of so-called “tsunami magnitudes” (Iida, 1963; Abe, 1981), which consists of compiling and averaging tsunami run-up heights at various sites, regardless of the often nonlinear response of specific coastal bathymetry and harbors. By conducting full-scale run-up simulations with specific earthquake-source scenarios, we can provide more realistic estimates of the actual tsunami hazard at a site such as Pisco. Our sources include the tsunami catalogs of Solov’ev and Go (1984) and Solov’ev et al. (1986), and the seismological compilations of Silgado (1992) and Dorbath et al. (1990). Other catalogs of South American tsunamis have been published, notably by Lomnitz (1970) and Lockridge Evaluation of Tsunami Risk from Regional Earthquakes at Pisco, Peru 1635 Figure 1. Map of central Peru detailing (inset) the location of the port of Pisco, Playa Loberia, and the proposed LNG terminal facility (star). On the main map, the color-keyed segments are models of rupture for historical tsunamigenic earthquakes. Triangles are epicenters of modern events from the instrumental era. For 1974 and 1868, dashed lines are alternate models involving a proposed longer rupture. To improve clarity, fault lines are traced at variable distances from the coast, but this is not meant to express differences in epicentral distance from the shoreline or in possible depth extent of the rupture. (1985), but all such compilations tend to be based on interpretations of the same original reports, and for this reason we refer, for each event and whenever possible, to those individual studies. In very general terms, we can classify tsunamis according to the amplitude of their run-up on the coast: decimetric tsunamis (with run-up between 0.1 m and 1 m) are mostly recorded by tide gauges and do not carry a specific hazard over and beyond that presented by storm waves; an example in Peru would be the Nazca earthquake of 12 November 1996 (Swenson and Beck, 1999). Metric tsunamis (with runup of a few meters) can inflict substantial damage to coastal and harbor communities, especially because they can result in inundation distances of up to 1 or 2 km inland, as demonstrated recently by the Camaná tsunami of 23 June 2001 (Okal et al., 2002). Finally, dekametric tsunamis (with runup greater than 10 m) are catastrophic events leading to the total destruction of coastal communities, with inundation reaching several kilometers inland. Examples would include the overflow of the saddle between Banda Aceh and Lhoknga over a total distance of 15 km during the recent 2004 Sumatra tsunami (Borrero, 2005), and the 1868 Arica, Peru (now Chile), tsunami (Billings, 1915), which deposited the U.S.S. Wateree 3.5 km from the shoreline. For the purpose of the present study, the first class of events are not considered to contribute to tsunami risk. Note that our ranking shares the philosophy of Solov’ev’s (1970) tsunami intensity scale while using a simplified three-tier classification. From the standpoint of the seismological investigation of the parent earthquakes, we can distinguish three periods: After 1976 (in the era of digital instrumentation), the Centroid Moment Tensor (CMT) project at Harvard University (Dziewonski et al., 1987 and subsequent quarterly updates) routinely computes a coherent set of earthquake source parameters, and the size of the rupturing fault can be inferred from source tomography and the relocation of aftershocks (e.g., Bilek and Ruff, 2002) During the period of instrumented historical seismicity (roughly from 1900 to 1976), seismic records available from several archival sites can be used to apply techniques similar to (or derived from) the CMT algorithm (Kanamori, 1970; Okal and Reymond, 2003) to retrieve the earthquake source characteristics; most large 1636 E. A. Okal, J. C. Borrero, and C. E. Synolakis earthquakes with tsunami potential have been the focus of individual studies (e.g., Swenson and Beck, 1996). In addition, epicentral locations are available, and precise relocations can be carried out (Wysession et al., 1991; Engdahl et al., 1998), using datasets of arrival times listed by the International Seismological Centre (ISC; earlier the International Seismological Summary or ISS). For events predating 1900, when no seismic records are available (preinstrumental era), the size of the largest earthquakes can often be inferred through the use of macroseismic techniques such as the interpretation of isoseismal lines, which in turn can be reconstructed from detailed reports of destruction by civilian authorities and clergy, for example, and compiled in Peru by Silgado (1992). For all practical purposes, and to our best knowledge, no quantifiable records exist for any seismic and/or tsunami disasters in Peru predating the Spanish colonization in 1531. In addition to their classical generation through the deformation of the seafloor under the coseismic dislocation, tsunamis damaging in the near field can also result from underwater landslides, themselves triggered during or shortly after a seismic event. This scenario, which was responsible for the disastrous 1998 tsunami in Papua New Guinea (Synolakis et al., 2002), should be kept in mind when assessing tsunami hazard, even though the quantification of its risk is made immensely difficult by the fundamentally nonlinear nature of the triggering phenomenon, and by our general ignorance of the population statistics of underwater landslides. Review of Historical Tsunamis Having Affected the Central and Southern Coasts of Peru In this section, we review systematically the largest earthquakes documented along the central and southern coasts of Peru over the past 350 years. We focus on the tsunamis they generated, especially regarding their impact on the city of Pisco, either as reported in various sources, or in the case of older events, as inferred from reports at other locations along the coast. For each significant tsunami, we either retrieve from the literature, or otherwise estimate, a source model that is later used in our numerical simulations. Events are listed chronologically, starting with the most recent. 23 June 2001, Camaná, M0 4. 9 10 28 dyne cm (Not Observed at Pisco) This is the most recent large tsunamigenic earthquake, with a human toll attributable to the tsunami of 22 fatalities and 52 missing. Reports from the International Tsunami Survey Team (Okal et al., 2002) indicate a maximum run-up of 7 m at Camaná; the tsunami was not observed above the high-water mark outside the Yauca-Ilo segment of the coast. We do not model this tsunami. 12 November 1996, Nazca, M0 4. 6 10 27 dyne cm (Not Reported at Pisco) This low-angle thrust event involved the subduction of the prominent Nazca Ridge, at a location recognized earlier as a seismic “gap” between the rupture zones of the 1942 and 1974 events (Beck and Nishenko, 1990). The tsunami was decimetric, reaching only 35 cm (peak-to-trough) at Arica and 24 cm at Callao. We do not model this tsunami. 21 February 1996, Chimbote “Tsunami Earthquake,” M0 2. 2 10 27 dyne cm (Not Reported at Pisco) This was a typical “tsunami earthquake” (Kanamori, 1972; Newman and Okal, 1998), characterized by a slow rupture at the source, resulting in enhanced tsunami excitation relative to its moment inferred from seismic mantle waves. The run-up reached 5.1 m at Chimbote, resulting in 12 deaths and considerable damage to more than 150 houses and beach huts (Bourgeois et al., 1999). However, it was not recorded above the high-water mark south of 11 S. We do not model this tsunami. 03 October 1974, Lima, M0 1. 5 10 28 dyne cm (Okal, 1992) (Mild Destruction at Pisco) The tsunami was recorded by maregraphs with peak-totrough amplitudes of 1.83 m at Callao and 1.2 m at San Juan. It reportedly inundated houses on the waterfront at Pisco, which would suggest amplitudes of at least 2 m, but less than 4 m, which would probably have led to more systematic destruction. This earthquake was studied by several authors, including G. S. Stewart (reported by Kanamori, 1977a), Dewey and Spence (1979), and in detail by Beck and Ruff (1989). Okal and Newman (2001) showed that its source was marginally slow. Based on the aftershock distribution, Dewey and Spence favored a 250-km-long fault, whereas the source tomography by Beck and Ruff suggested a shorter length. Consequently, we test two models, a short fault (model a) extending 150 km with a slip of 5 m, and a long fault (model b) extending 250 km, but with a slip of only 3 m, the latter shown as a dashed line on Figure 1. We use a hypocentral depth of 22 km, and Stewart’s focal mechanism ( 340 ; d 17 ; k 90 ). 03 September 1967, M0 6. 3 10 26 dyne cm (Okal and Newman, 2001) (Not Reported at Pisco) This earthquake generated only a minor tsunami with decimetric amplitudes at Callao and Chimbote. It is clearly too small to bear significantly on the tsunami risk along the coast, and we do not model it. 17 October 1966, Barranca, M0 1.95 10 28 dyne cm (Abe, 1972) (Not Reported at Pisco) The tsunami reached 6 m at Tortuga (presumably on the northern coast) and a peak-to-trough amplitude of 3.5 m at Evaluation of Tsunami Risk from Regional Earthquakes at Pisco, Peru 1637 Callao, and was recorded at San Juan (no amplitude reported). Thus, it would be feasible that a metric oscillation took place at Pisco. The earthquake was studied by Abe (1972), Dewey and Spence (1979), Beck and Ruff (1989), and Okal and Newman (2001). We model the rupture as a 150-km-long, 50km-wide fault with a slip of 4 m. The focal mechanism ( 330 ; d 12 ; k 90 ) is taken from Abe’s work, the hypocentral depth (38 km) from the ISC pP determination. 24 August 1942, San Juan, M0 1.3 10 28 dyne cm (Okal, 1992) (Not Reported at Pisco) This large earthquake was centered east of the Nazca Ridge, but a relocation by Okal and Newman (2001) suggests that its epicenter was on land. This could explain that despite the large seismic moment, tsunami damage was limited, with a lone report at Lomas; indeed Solov’ev and Go (1984) suggested that the tsunami may have been generated by a local landslide. We do not model this event. 24 May 1940, Huacho, M0 2.5 10 28 dyne cm (Kanamori, 1977a; Okal, 1992) (Possibly Recorded at Pisco) The tsunami is poorly documented, but reports quoted by Solov’ev and Go (1984) describe it as stronger than in 1966. The earthquake was studied by Beck and Ruff (1989) and Okal and Newman (2001). We model the fault as 50km-wide, 120-km-long fault with a slip of 5 m. The hypocentral depth is taken as 40 km, and the focal mechanism after Beck and Ruff ( 340 ; d 17 ; k 90 ). 13 August 1868, Arica, M0 7 to 10 10 29 dyne cm (Estimated) (Total Destruction at Pisco) This is the great Arica tsunami, probably the largest to affect Peru in historical times. Although no run-up amplitude is available for Pisco, Solov’ev and Go (1984) reported that “[the tsunami] destroy[ed] everything in its path. In particular, a stone breakwater was utterly flattened.” Catastrophic destruction was also reported in the nearby Chincha Islands, where a down-draw of at least 20 m was observed. Run-up heights of 15 m or more were reported along the Southern Coast, for example, in Mejia. The tsunami was recorded all over the Pacific, including in Japan and New Zealand with amplitudes of several meters, a situation repeated in modern times only by the 1960 Chilean earthquake in the Pacific Basin and the 2004 Sumatra earthquake in the Indian Ocean. Thus, it seems warranted to give this event an exceptionally large size. The detailed investigation of isoseismals by Dorbath et al. (1990) suggests a length of rupture of 600 km, which would be surpassed among modern events only by Chile (1960), Alaska (1964), and Sumatra (2004). In this framework, we tentatively assign the Arica earthquake a moment of 7 10 dyne cm (i.e., slightly smaller than the estimate for the 1964 Alaska earthquake using conventional mantle waves [Kanamori, 1970]), and model its rupture as a 600-km-long, 150-km-wide fault with a slip of 15 m. Our chosen focal mechanism ( 305 ; d 20 ; k 90 ) expresses the local geometry of subduction. As discussed under “Hydrodynamic Simulations,” we also consider an even larger source, by prolonging the rupture 300 km to the northwest in the azimuth 330 . This would imply rupturing through the Nazca Ridge segment and boost the seismic moment to 10 dyne cm.