1 Introduction

It has now been slightly over a quarter century since two very different earthquakes literally re-awakened the field of observational tsunami science, which had been to some extent stagnant since the catastrophic events of the 1950s and early 1960s (Kamchatka, 1952; Aleutian, 1957, Chile, 1960; Alaska, 1964), despite significant tsunamis in 1976 in Mindanao, Philippines, and in 1983 in the Sea of Japan. While substantial progress was made in the 1970s and 1980s on the theoretical and experimental front (e.g., Hammack 1973; Ward 1980; Bernard and Milburn 1985), scientists still lacked the motivation provided by exceptional and intriguing field observations.

The two tsunamis of 02 September 1992 in Nicaragua and 12 December 1992 in Flores Island, Indonesia provided our community with a wealth of challenges which spawned a large number of new investigations covering the observational, experimental and theoretical aspects of our science. They were followed over the next 25 years by many more events (in particular by the catastrophic tsunamis of 2004 in Sumatra, and 2011 in Japan), many of which resulted in significant new scientific developments, which are the subject of this special issue.

We restrict our study to those “geological” tsunamis generated by earthquakes, landslides and volcanic eruptions. Whether their origin rests with strain accumulated during the interseismic cycle (and released during earthquakes), gravitational unbalance resulting from geological processes (landslides), or volcanic activity, their energy can be traced eventually to convective heat loss in the Earth’s interior in the context of Plate Tectonics. By contrast, a separate paper (Rabinovich 2019) examines the case of “meteo-tsunamis” generated by coupling between atmospheric perturbations and the oceanic column. As such, the energy of meteo-tsunamis is drawn from the weather system, and eventually from the radiation received from the Sun.

In this paper, we present a detailed, chronological discussion of the unique properties of a selection of 47 among such geological tsunamis. Their sources are mapped on Fig. 1 and their fundamental parameters listed in Table 1. We emphasize that the list is not exhaustive, and represents a choice of events of significance in the development of tsunami science, out of the 320 events listed for that time window in NOAA’s Global Tsunami database, 39 of which involving casualties, as reviewed in the companion paper by Gusiakov et al. (2019). Like all such choices, it may be subjective. Chronological plots of tsunami casualties and maximum run-up are given on Figs. 2 and 3. In this context, both Table 1 and Fig. 2 attempt to reflect tsunami casualties, not counting those directly attributable to the parent earthquake, which in some instances (e.g., İzmit, Event 13 or Pisco, Event 26) were much heavier; we stress that in most cases, such a distinction remains a difficult exercise. Figure 4 shows energy-to-moment ratios (Newman and Okal 1998) for the 39 relevant earthquake sources.

Fig. 1
figure 1

Map of the 47 events described in the text. Special symbols are used for tsunami earthquakes, for the deep Okhotsk earthquake (37), for tsunamis generated during volcanic events, and for those primarily due to major landslides. Event numbers are keyed to Table 1 and the text

Full size image
Fig. 2
figure 2

History of the 47 events (1992–2018). The vertical bars are scaled to the logarithm of the death toll, with events without known casualties plotted below the dashed line. Symbols are keyed to Fig. 1

Full size image
Fig. 3
figure 3

History of the 47 events (1992–2018). The vertical bars are scaled to the logarithm of the maximum observed run-up. The deep Okhotsk event (Number 37), for which no data is available, is plotted below the dashed line. Symbols are keyed to Fig. 1

Full size image
Fig. 4
figure 4

Energy-to-moment ratios for the 39 earthquake sources among the events studied (1992–2018). This plot is analogous to those published, e.g., by Newman and Okal (1998). The dashed lines identify constant values of \(\Theta \), with the grey solid line corresponding to the theoretical value (\(-\,4.90\)) expected from scaling laws. Tsunami earthquakes (featuring \(\Theta \, \le -\,5.8\)) are shown as red stars and keyed to their number in Table 1. Events defined as “snappy” (Okal and Saloor 2017) (\(\Theta \, \ge -\,4.7\)) are similarly shown as purple diamonds, with the deep Sea of Okhotsk earthquake shown in yellow

Full size image

2 Significant Tsunamis, 1992–2018: A Timeline

1. :

Nicaragua, 02 September 1992

This event was remarkable in many respects. First, it constituted the first “tsunami earthquake” to take place in the era of digital seismic instrumentation. We recall that the term “tsunami earthquake” was coined by Kanamori (1972) to characterize an event whose tsunami is significantly larger than expected from its magnitudes, especially conventional ones. Such events are particularly ominous as they can be felt, if at all, at a deceptively low level of shaking, while the ensuing tsunami can be catastrophic (170 fatalities in Nicaragua), thus negating the adage “The Shaking is the Warning”, commonly used to sensitize populations at risk on tsunami-prone coastlines in the near field. While Kanamori (1972) was able to suggest that their large tsunamis resulted from an anomalously slow seismic rupture, the charter tsunami earthquakes in his study, the 1896 Sanriku and 1946 Aleutian shocks, obviously predated digital seismology. It was at the time extremely challenging if not outright impossible to quantify the high-frequency part of the seismic spectrum from analog records. Following the deployment of broad-band digital instruments (Hutt et al. 2002), Boatwright and Choy (1986) made this possible by developing an algorithm to compute radiated seismic energy, and the 1992 Nicaragua tsunami earthquake (together with similar ones in 1994 in Java and Chimbote, Peru in 1996) motivated Newman and Okal (1998) to introduce the source slowness parameter \(\Theta = \log _{10} (E^{E} / M_0 )\) where \(E^{E}\) is the estimated radiated seismic energy and \(M_0\) the seismic moment, as a robust identifier of tsunami earthquakes.

Table 1 Significant tsunamis, 1992–2018
Full size table

Second, the availability of high-quality broad-band data allowed Kanamori (1993) to discover, on records of the Nicaraguan earthquake, an ultra-long period, fast-propagating seismic wavetrain, which he named the W phase, with the potential for a rapid estimate of the long-period seismic moment controlling the generation of tsunamis. Despite some initial interest (Okal 1993), the concept had to wait for the full development of digital networks to allow its now routine use for rapid, full-fledge moment tensor inversion (Kanamori and Rivera 2008).

Next, the 1992 Nicaragua tsunami saw the first of comprehensive International Post-Tsunami Surveys (Abe et al. 1993), which were to become systematic following later significant events (Synolakis and Okal 2005), even though similar efforts had occasionally taken place, notably after the 1976 Mindanao tsunami (Wallace et al. 1977). The substantial values of run-up documented by the 1992 survey (8–10 m) could not be modeled using then standard simulation algorithms, which stopped the simulation at a shallow, but arbitrary, water depth (typically 5–10 m), and considered the coastline as a fully reflecting boundary (Imamura et al. 1993), thus underestimating the surveyed values by as much as one order of magnitude. This motivated the development of codes extending the computation over initially dry land (Titov and Synolakis 1993), which later matured into codes such as, e.g., MOST (Titov and Synolakis 1998; Titov et al. 2016), TUNAMI–2 (Goto et al. 1997), or COMCOT (Liu et al. 1994).

Finally, the 1992 Nicaragua event provided the first opportunity for space-based detection of a tsunami on the high seas, using an 8-cm signal on ERS-1 and TOPEX/POSEIDON radar altimeters (Okal et al. 1999).

2. :

Flores, 12 December 1992

This event took place on the Northern shore of Flores Island, Indonesia, where the Sunda plate subducts Southwards under the Australian plate. The tsunami was remarkable on two accounts. First, it gave rise to the intriguing “Babi effect”: at Babi Island, located a few km North of the main island, the maximum run-up (7.2 m) occurred in the lee of the incident tsunami, as the waves interfered constructively after rounding this circular island (Yeh et al. 1993). This effect was later reproduced in the laboratory (Briggs et al. 1995) and modeled theoretically (Yeh et al. 1994; Liu et al. 1995; Imamura et al. 1995a; Tinti and Vannini 1995). It negated the conventional wisdom suggesting that island structures could protect their residents on the lee side of an incoming tsunami wave.

In addition, the 1992 Flores tsunami featured an isolated extreme run-up of 26 m at Rangrioko in Northeastern Flores (Tsuji et al. 1995), which underwater surveys later documented was caused by a submarine landslide triggered by the earthquake (Plafker 1997). This phenomenon contributed to raising awareness for this particular kind of hazard.

3. :

Okushiri, 12 July 1993

This event took place in the Sea of Japan and resulted in the destruction of the port village of Aonae, with a run-up of 10 m. A detailed field survey (Shuto and Matsutomi 1995) documented a run-up reaching 31 m due to amplification by a river valley at the so-called “Tsuji Gulch”. Later simulation by Titov and Synolakis (1997) established the possibility of successfully modeling the highly non-linear interaction of tsunamis with originally dry land featuring extreme relief.

4. :

East Java, 02 June 1994

This was the second tsunami earthquake of the 1990s, running up as much as 16 m on beaches where the earthquake had not been felt, and resulting in 238 deaths (Synolakis et al. 1995).

5. :

Shikotan, Kuriles, 04 October 1994

This large earthquake (at the time the second largest in the CMT catalog) represented a tear in the slab below the Southern Kuriles. It resulted in significant local destruction, but the tsunami inflicted no casualties despite run-up of 5–9 m, as surveyed by Yeh et al. (1995). It nevertheless emphasized the need to include events other than interplate thrusts in the assessment of tsunami hazard at subduction zones, as already evidenced by such disasters as the 1933 Sanriku tsunami (Okal et al. 2016).

6. :

Skagway, Alaska, 03 November 1994

This local tsunami in the port of Skagway destroyed a wharf in a shipping terminal and killed one worker, in the absence of any detectable earthquake (Kulikov et al. 1996). Thomson et al. (2001) suggested that the failure of the wharf was responsible for the tsunami; however, the identification of an underwater slide down the fjord led Synolakis et al. (2002a) to a successful simulation of the failure of the wharf under a system of waves generated by that natural landslide.

7. :

Mindoro, Philippines, 14 November 1994

This was a very significant tsunami, which ran up as high as 7.3 m, killed 81 persons, and displaced a 4000-ton barge 1.6 km in a local estuary (Imamura et al. 1995b). Yet, the parent earthquake had a strike-slip mechanism, generally perceived as posing little if any tsunami risk. However, Tanioka and Satake (1996) showed that the lateral offset produced by a strike-slip fault at the edge of the island can generate a tsunami in the same fashion as a wavemaker paddle in the laboratory. The reverse situation (a marine strike-slip fault hitting the head of a bay) is thought to be responsible for the catastrophic 2018 tsunami at Palu, Indonesia (see 46. below), even though ancillary landslides certainly played an additional role in its generation (Satake, pers. comm., 2019).

8. :

Manzanillo, Mexico, 09 October 1995

This was a typical tsunami generated by a major interplate thrust earthquake at the Mexican subduction zone. As part of the field survey, Borrero et al. (1997) obtained photographs from witnesses documenting a leading recession at Tencatita Bay. Although this phenomenon had been predicted theoretically by Tadepalli and Synolakis (1994, 1996), this observation verified that, in the most widely expected geometry of a shallow angle interplate thrust fault, the local beach will benefit from a natural warning to the population at risk, in the form of an initial down-draw; this may help explain the relatively low death toll, reported between 1 and 10.

9. :

Biak, Indonesia, 17 February 1996

This tsunami took place at a subduction zone featuring very oblique convergence, resulting in plate motion partitioning, and in a scarcity of megathrust earthquakes at the subduction interface, a situation to be repeated during the 2012 Haida Gwaii earthquake (see 35. below). A remarkable aspect of the field survey is the estimation, for the first time, and based on an application of Bernoulli’s theorem, of current velocities (Matsutomi and Iizuka 1998; Matsutomi et al. 2001).

10. :

Chimbote, Peru, 21 February 1996

Just 4 days after the Biak event, a new significant tsunami took place in Northern Peru, following the third “tsunami earthquake” of the decade. Investigations into the anomalous slowness of the earthquake (Ihmlé et al. 1998) led to successful tsunami simulation using for the first time a reduced rigidity \(\mu \) for the seismic source material (Heinrich et al. 1998a).

11. :

Montserrat, Lesser Antilles, 26 December 1997

This tsunami resulted from the penetration of the ocean by a pyroclastic flow during the eruption of Montserrat. It ran up 1 m at Deshaies on the nearby island of Guadeloupe, in a geometry comparable to that of the 2018 Krakatau disaster (see 47. below), but on a much reduced scale. Remarkably, this scenario had been accurately anticipatedFootnote 1 by Heinrich et al. (1998b). It was essentially repeated at the same location on 13 July 2003 (Pelinovsky et al. 2004).

12. :

Aitape, Papua New Guinea, 17 July 1998

With about 2100 casualties, this tsunami was one of the deadliest of the past century. Yet, it followed a moderate earthquake whose moment \((M_0\,=\,3.7\,\times \,10^{26}\)\({\hbox {dyn}}\,\times \,{\hbox {cm}}\)) could not justify run-up reaching 15 m on a narrow stretch of coastline centered on Sissano Lagoon (Kawata et al. 1999). The combination of extreme amplitudes, limited lateral extent of the devastation, minimal amplitudes in the far field, and an apparent delay in the arrival of the wave, led Synolakis et al. (2002b) to propose as the origin of the tsunami a submarine landslide triggered 13 minutes after the seismic shock. The landslide was later identified during a survey involving seismic reflection and submarine ROV operations (Sweet and Silver 2003), and documented through its T phases throughout the Pacific Basin (Okal 2003).

The high death toll of the Papua New Guinea tsunami raised awareness among scientists and Civil Defense officials worldwide for hazards from potential landslides, notably in areas where offshore seismicity would have otherwise minimized tsunami risk, such as the Los Angeles Basin (Borrero et al. 2003).

13. :

Kocaeli (İzmit), Turkey, 17 August 1999

This tsunami added significant damage in the Eastern part of the Sea of Marmara to a catastrophic earthquake with a total death toll of 17,000. Its source was a combination of a strike-slip mechanism at the head of İzmit Bay (in retrospect comparable to the 2018 Palu geometry (see 46. below)), vertical motion at pull-apart basins, and possible landslides in the Sea of Marmara, with inundation aggravated by local subsidence, notably at Gölçuk (Altınok et al. 2001; Tinti et al. 2006).

This event is particularly significant, as it falls in the pattern of inexorable, if somewhat slow and certainly irregular, westward stress transfer along the Northern Anatolian Fault, initiated in Erzincan in 1939 (Stein et al. 1997). As such, it casts the ominous augury of the next event, expected to occur off İstanbul, a megalopolis with a population of 15 million.

14. :

Fatu Hiva, Marquesas, 13 September 1999

In the absence of any detectable seismic tremor, this tsunami was caused by the spontaneous collapse, into the sea, of a 4 million \(\mathrm{m}^3\) rock slide, part of the flank of an eroding 1.3 m.y. old volcano, following a 4-month drought throughout the Pacific, which acted to desiccate clays resulting from alteration of volcanic dykes (Okal et al. 2002a). It destroyed coastal infrastructure, including the wall of the school yard of the village of Omoa, and flooded the school where miraculously, 85 children escaped unharmed.

In a rare example of proactive development, the village school was relocated out of harm’s way a few years later, about 1 km up the valley.

15. :

Vanuatu, 26 November 1999

The tsunami followed a moderate earthquake (Pelletier et al. 2000), and is noteworthy because of the successful evacuation of the village of Baie Martelli on the Southern coast of Pentecost Island (Caminade et al. 2000). The villagers had been sensitized to tsunami danger through a video based on the 1998 Fatu Hiva disaster, stressing the need to immediately self-evacuate low-lying areas upon feeling strong earthquake tremors, especially if accompanied by a recess of the sea. When the earthquake struck in the middle of the night, and a down-draw was reported, the village chief ordered an immediate full evacuation. The tsunami completely destroyed the village, but, out of the 300 residents, only three lost their lives: two elderly invalids who could not be evacuated and a drunken man, who refused to.

The lesson, if need be, from this event is simply: “Education works!”

16. :

Camaná, Peru, 23 June 2001

The earthquake was, at the time, the largest recorded worldwide in 36 years, and it generated a tsunami damaging 400 km of coastline, and killing 86 people (Okal et al. 2002b). A most remarkable aspect of the field survey was the concentration of the casualties on the river delta at Camaná, where farmers from the hinterland, poorly sensitized to tsunami hazards, were at work, while residents of fishing communities with a strong ancestral tradition had largely self-evacuated upon noticing the initial down-draw (Dengler et al. 2003). A much greater disaster was avoided by the occurrence of the tsunami on a beach deserted during the Winter season, which would have been packed with tourists during the Summer.

The Camaná event is also remarkable in that it provided the first detection of a tsunami through the oscillation of the ionosphere due to the continuation of its eigenfunction into the atmosphere (Artru et al. 2005), an idea originally suggested by Peltier and Hines (1976).

17. :

Stromboli, Italy, 30 December 2002

This volcanic tsunami was triggered by landslides, one submarine, the other aerial falling into the sea, at La Sciara on the flank of Stromboli (Tinti et al. 2005). It caused major damage in the nearby villages of Piscità and Ficogrande, but again a major disaster was avoided thanks to its occurrence during a season of low tourist activity.

A remarkable aspect of this event was the recording of the actual tsunami waves on horizontal seismometers located 20 km away on the island of Panarea (La Rocca et al. 2004), to our knowledge the first modern such instance, following Angenheister’s (1920) observation in the far field after the Kuriles event of 07 September 1918.

18. :

Aleutian Islands, 17 November 2003

While this tsunami caused no damage or casualties, and had a maximum reported run-up of only 26 cm at Shemya, it is noteworthy because it constituted the first use of DART sensor records as part of a real-time accurate forecast of tsunami amplitudes inside a distant harbor, in this case 20 cm at Hilo, Hawaii (Titov et al. 2005a). The alert in the Hawaiian Islands was canceled, thus preventing an unnecessary evacuation, and saving an estimated 68 million USD (Meining et al. 2005).

19. :

Sumatra-Andaman, 26 December 2004

With more than 200,000 casualties, this tsunami, the first in recorded history to have exported death and destruction across the Indian Ocean Basin, was probably the most lethal one in the history of mankind, considering the exponential growth of coastal populations in the past few centuries. The tsunami was the first in the era of modern technology to be detected worldwide (Rabinovich and Thomson 2007) and to be simulated on a full spherical model of the Earth’s oceans (Titov et al. 2005b). It transformed the word “tsunami”, used at the time only by a small community of specialists, into a worldwide household term.

It cannot be our purpose here to list every scientific result obtained in its wake, but rather we pinpoint fundamental observations which had a seminal effect on tsunami science, technology or societal response in the following years.

  • From the operational standpoint, the failure to provide adequate warning to distant populations was traced to inexistent means of communication and unpreparedness in the coutries at risk, rather than to a systematic deficiency in scientific analysis (Marris 2005); at the time, the existing warning centers, in particular PTWC and ATWC, had no responsibility for the Indian Ocean, and hence no way to transmit a warning following their detection of the earthquake. The 15 years elapsed since the Sumatra disaster have seen a considerable investment into the worldwide development of a number of tsunami warning centers, notably in the Indian Ocean and in particular in Australia, Indonesia, and Oman. Only the trial of time will evaluate their performance for far-field warning.

  • Perhaps the most important scientific result in the wake of the 2004 tsunami was the development of a new CMT inversion algorithm based on the W phase (Kanamori and Rivera 2008), following its observation at a record amplitude during the 2004 earthquake (Lockwood and Kanamori 2006). The algorithm has now been implemented worldwide at warning centers (Hayes et al. 2009), and has become the new de facto standard in long-period moment-tensor inversion.

  • Another crucial development has been the abandonment of the paradigm linking plate age and convergence rate to a proposed maximum earthquake magnitude at subduction zones, originally proposed by Ruff and Kanamori (1980). Under this model, the Sumatra trench should not have hosted events bigger than \(M_w \approx 8.2\), below the threshold for transoceanic tsunami hazard. Noting that a similar problem plagued the Cascadia subduction zone, where evidence for the 1700 megathrust event (Satake et al. 1996) came posterior to Ruff and Kanamori’s (1980) study, Stein and Okal (2007) conducted a critical analysis of an updated dataset of tectonic and earthquake parameters, finding a significantly degraded correlation to support the paradigm; they also noted that Ruff’s (1989) later suggestion to involve trench sediment thickness fared no better. In this context, a precautionary attitude is to consider that any sufficiently long subduction zone may be the site of a mega-event in the future, a conclusion shared by McCaffrey (2007).

  • A large number of observations established the slowness of the seismic rupture of the 2004 event. They ranged from the growth of seismic moment with period (Stein and Okal 2005), seismic source tomography (e.g., Ishii et al. 2005), and the duration of high-frequency P waves (Ni et al. 2005) and hydroacoustic T waves (Guilbert et al. 2005), in addition to a clearly deficient \(\Theta \) parameter (Stein and Okal 2007). In this context, and following Choy and Boatwright (2007), we classify the event as a “tsunami earthquake” on Figs. 1, 2, 3 and 4.

    In view of the slow precursor to the 1960 Chilean earthquake identified by Kanamori and Cipar (1974), and of the reassessment of the low-frequency moment of the 1964 Alaskan event (Nettles et al. 2005), this slowness in the source of the 2004 earthquake raises the legitimate question of whether all “super-mega” earthquakes (\(M_0 > 10^{30}\)\({\hbox {dyn}}\,\times \,{\hbox {cm}}\)) feature a similar behavior, which would have significant implications for tsunami hazard in their aftermath (Okal 2013); the extreme scarcity of such events (a grand total of 3 have been analyzed) leaves the jury out at this time.

  • The 2004 Sumatra earthquake was so large that its tsunami was recorded by a number of instruments which were not designed for it, generally as the result of coupling between very different media, such as the sea, the solid Earth and the atmosphere. This coupling would be expected to be very small, but in the presence of a monumental event, can yield tangible observations which were not only documented, but also often quantitatively forward- or inverse-modeled, occasionally all the way to an appropriate estimate of the moment of the earthquake. This was the case, for example of hydrophones of the IMS (Hanson and Bowman 2005; Okal et al. 2007), radar altimeters (Smith et al. 2005) leading to an explanation of the intriguing case of “tsunami shadows” (Godin et al. 2009), of the detection of the tsunami in the ionosphere from perturbation to GPS signals (Liu et al. 2006; Occhipinti et al. 2006), and systematically by horizontal seismometers on oceanic islands (Yuan et al. 2005) in a pattern that Okal (2007) showed could be interpreted as mirroring an Ocean-Bottom Seismometer record. Most such techniques were later applied regularly to smaller tsunamis.

    Finally, the tsunami was recorded by seismometers temporarily deployed on large icebergs floating off the coast of Antarctica, providing what remains to this day the only in situ recording of the horizontal motion of the ocean surface during the passage of a tsunami (Okal and MacAyeal 2006).

  • Surveying in the near field benefited from technological advances such as the scientific decoding of video recordings to reconstruct a quantitative estimate of tsunami flow velocities (Fritz et al. 2006), which in particular documented their increase with time within a wavetrain, leaving essentially no chance for the population to escape.

  • An unexpected observation compiled at several locations during the numerous field surveys covering the entire Indian Ocean Basin, was the occasional development of strong currents in distant harbors several hours after the arrival of the initial wave, and often after the all clear had been given by local authorities, e.g., in Toamasina, Madagascar (Okal et al. 2006a), Le Port, Réunion (Okal et al. 2006b) and Salallah, Oman (Okal et al. 2006c). This effect, which can lead to hazardous situations when major vessels break their moorings in ports, is explained as the setting up of harbor resonance by higher-frequency components of the tsunami traveling outside the shallow-water approximation. Since 2004, similar effects have been observed during other events. In this context, harbor communities must be educated to this form of hazard, and evacuation procedures, especially the issuance of all clear, adapted accordingly.

20. :

Nias, Indonesia, 28 March 2005

This “second” Sumatra event, remarkably anticipated a few days earlier by McCloskey et al. (2005) in the context of Coulomb stress transfer, was indeed gigantic (\(M_0 = 1.05 \times 10^{29}\)\({\hbox {dyn}}\,\times \,{\hbox {cm}}\)), and took 600 lives in Nias and 100 in Simeulue. However, its tsunami was modest, running up to only 4.2 m, apparently without fatalities in the near field. It was also negligible in the far field, despite 10 casualties in traffic accidents due to panic generated by night-time evacuation (Okal et al. 2006a; Borrero et al. 2011). This intriguing situation is explained by a significant fraction of the deformation in the source area taking place either under large islands (“the earthquake moved a large amount of rock and not much water”), or at shallow depths, the tsunami then faltering under Green’s (1838) law when reaching deep basins (Synolakis and Arcas, quoted by Kerr 2005).

In addition, the population of Simeulue was sensitized to tsunami hazard through the Smong tradition, possibly derived from the catastrophic “tsunami earthquake” of 1907 (Martin et al. 2019), and self-evacuated during both the 2004 and 2005 tsunamis (McAdoo et al. 2006).

21. :

Central Java, 17 July 2006

This “tsunami earthquake” was essentially a repeat of the 1994 event (see 4. above), 580 km to the East. As such, it reopens the question of any possibly regional character to tsunami earthquakes. In an early study, Okal and Newman (2001) had found meager evidence for it in the Peruvian system. The 2006 event in Java shows a stronger case, supported also by the 2012 El Salvador and 2013 Santa Cruz events (see 34. and 36. below).

22. :

Central Kuriles, 15 November 2006

23. :

Central Kuriles, 13 January 2007

We discuss those two events together, since they occurred as a seismic doublet, in Winter, in a remote, unpopulated part of the island chain, and it took 9 months before a field survey could be arranged (MacInnes et al. 2009), which thus could not assign a priori the field evidence (up to 22 m of run-up) to one or the other. Tsunami records in the far field (at a minimum distance of 600 km) were obtained at tidal gauges across the Pacific and on DART sensors (Rabinovich et al. 2008), showing a smaller tsunami in 2007 at all locations.

The first event was a classical interplate thrust, followed 2 months later by a slightly smaller normal faulting event on the outer rise, probably triggered by stress transfer (Ammon 2008). As expected under such a scenario, source tomography studies (e.g., Lay et al. 2009) revealed that the smaller 2007 event featured larger slip distributed over a smaller surface than the 2006 mainshock. Using these source models, Rabinovich et al. (2008) carried out simulations which confirmed far-field observations, but suggested that the 2007 event could have featured larger amplitudes in deep water along the Central Kuriles than the 2006 one (on the order of 7 m as opposed to 5). Based on these results, and on full run-up calculations at a number of surveyed sites, MacInnes (2010) assigned some of her maximum run-up amplitudes to the 2007 event.

These two tsunamis emphasized the fundamental difference between near-field amplitudes, directly related to maximum slip on the fault, and far-field ones, controlled by the full integral of its distribution, i.e., by the seismic moment.

Perhaps because of the lack of local information, the warning centers for the Pacific Basin underestimated the 2006 event and canceled initial warnings, even though the tsunami was to cause one injury and damage in Hawaii, and especially in Crescent City, California, to the extent of 9.8 million USD (Dengler et al. 2009). The 2006 tsunami served as a particularly acute reminder of the vulnerability of that community to a wide range of tsunami sources in the Pacific, and resulted in the implementation of a more detailed, quantified set of warning criteria at the US Tsunami Warning Centers (Whitmore et al. 2008).

24. :

Gizo, Solomon Islands, 01 April 2007

This earthquake was remarkable in that it ruptured along the Solomon Islands trench, across the triple junction separating the Australian plate from the Woodlark mini-plate (Taylor et al. 2008), an important observation which shows that plate boundaries cannot be considered as barriers limiting the size of potentially tsunamigenic earthquakes in subduction zones. A similar scenario was proposed for the 1932 Manzanillo, Mexico earthquake by Okal and Borrero (2011).

The death toll (52) was contained despite widespread flooding and destruction, thanks to systematic self-evacuation in this region characterized by a high level of seismic and volcanic activity (Fritz and Kalligeris 2008). Very similar conclusions were reached after the tsunami of 03 January 2010, only 200 km to the ESE, which involved no casualties (Newman et al. 2011a).

25. :

Aysén, Chile, 21 April 2007

This tsunami resulted from a massive rock slide into Aysén Fjord (Barrientos et al. 2009), triggered by an earthquake on the nearby Liquiñe-Ofqui fault system (LOFS), which marked the culmination of an intense seismic swarm lasting 3 months (Legrand et al. 2011). The remoteness of the area helped contain the death toll, but splashes on Mentirosa Island reached \(\sim \)50 m, in a scenario reminiscent of the 1958 Lituya Bay tsunami (Miller 1960), albeit on a smaller scale. The exact role of the LOFS in the local plate tectonic framework should be further investigated in the context of local tsunami hazard, especially since Kanamori et al. (2019) have recently suggested that it may have contributed a significant component to the great 1960 Chilean earthquake.

26. :

Pisco, Peru, 15 August 2007

Following heavy structural damage and casualties inflicted by the earthquake on this Central Peruvian city, the tsunami locally reached run-up heights of 10 m, but “sergeants” acting in a sense as lifeguards on beaches conducted an evacuation that helped contain the death toll from the tsunami to three casualties in the town of Lagunilla, which inexplicably was not part of the program (Fritz et al. 2008).

It is noteworthy that this event had been anticipated, albeit on a greater scale, by Okal et al. (2006d).

27. :

Bengkulu, Indonesia, 12 September 2007

This very large event produced a relatively contained tsunami, with run-up not exceeding 4 m (Borrero et al. 2009), and despite significant tsunami damage, no deaths were reported, thanks to successful self-evacuation.

The event is noteworthy, since it took place in a region highlighted for tectonic strain accumulated since the 1833 mega-earthquake, as well as under stress transfer in the wake of the 2004 Sumatra and 2005 Nias earthquakes (Nalbant et al. 2005; Pollitz et al. 2006). However, the 2007 Bengkulu earthquake failed to release all the slip accumulated since 1833, and thus the area remains a candidate for a significant tsunami in the near future (Okal et al. 2009).

28. :

Chebalis Lake, B.C., 04 December 2007

This event resulted from the fall of a 3-million \(\mathrm{m}^3\) rockslide into Lake Chebalis in Southwestern British Columbia. Fortunately, the local campsites were closed for the Winter, and no lives were lost, but a wave splashed 38 m onto the opposite bank of the 1-km wide lake. The event was successfully modeled by Wang et al. (2015).

29. :

Samoa, 29 September 2009

This large event at the Northern corner of the Tonga arc actually had a composite source, starting as a normal faulting shock, followed \(\sim \)100 s later by a more regular interplate thrust one (Li et al. 2009; Lay et al. 2010). It raised a powerful tsunami, running up to 22 m on Tafahi, Tonga, and bringing death and destruction to American Samoa, Samoa, and Tonga, with a total of 192 casualties, mostly in [independent] Samoa.

Despite the complex mechanism, tsunami modeling based on the inversion of DART data assuming a regular shallow angle thrust provided acceptable fits to surveyed amplitudes, thus demonstrating the robustness of the approach (Okal et al. 2010). The field survey revealed a generally successful evacuation on American Samoa, reflecting a well-prepared community, as opposed to [independent] Samoa, where in particular the absence of signage may have contributed to a higher death toll (Fritz et al. 2011a).

30. :

Haiti, 13 January 2010

This is another example of a catastrophic strike-slip earthquake generating a minor tsunami, constrained to a few locations in Gonâve Bay and along the Southern coast (Fritz et al. 2012), with a run-up not exceeding 3 m, and a death toll of 7 (as opposed to perhaps 300,000 from the earthquake). At both locations, the tsunamis were due to underwater landslides triggered by the earthquake; note in particular that in this case, the Enriquillo strike-slip fault does not intersect the shoreline, preventing a “Mindoro”-type generation, which could have been much more damaging.

31. :

Chile, 27 February 2010

This megathrust event generated the first tsunami in 46 years with transpacific runup in excess of 1 m, in the Kuril Islands and at certain sites in California. In the near field, it caused considerable damage, with run-up reaching 29 m at Constitución, and 156 fatalities, many of them trapped on a campground on Orrego Island in the mouth of the Maule River (Fritz et al. 2011b). The real-time handling of the warning by Chilean authorities was generally regarded at best as deficient, at worst as inept, and in particular no warning was issued for Juan Fernández Island, where the loss of 18 lives could have been prevented, when the tsunami attacked the island 49 minutes after the earthquake. This led to a successful major restructuring of the Chilean Tsunami Warning Center (see 39. below).

The 2010 Maule tsunami was also the first one to be detected as a perturbation of the geomagnetic field of the Earth (Manoj et al. 2011), following Tyler’s (2005) model of induction currents caused by the displacement of conducting seawater in the Earth’s field.

32. :

Mentawai, Indonesia, 25 October 2010

This “tsunami earthquake” occurred up-dip of the 2007 Bengkulu event (see 27. above), in the typical configuration of an “Aftershock Tsunami Earthquake” (Okal and Saloor 2017). The high death toll (431) resulted from the combination of deceptively low shaking and of the occurrence, only a year before, of the intermediate-depth, intraplate Padang earthquake (Hill et al. 2012). The 2009 event had created a negligible tsunami (maximum run-up 27 cm), but had been strongly felt in the Mentawai Islands, instilling in the residents the false sense that a dangerous tsunami would require even greater shaking.

Incidentally, the anomalously slow nature of the earthquake had been documented 17 minutes after origin time (and thus before the tsunami could reach the Mentawai Islands), on the basis of the comparison of duration and energy of teleseismic P waves (Newman et al. 2011b), but there existed no way to transmit this information to Indonesia, let alone to the people at risk, many of whom were in communities lacking electrical power, where the local warning from the Indonesian Tsunami Warning Center was apparently not received.

33. :

Tohoku, Japan, 11 March 2011

The catastrophic 2011 Tohoku tsunami ranks only second the 2004 Sumatra one in terms of death toll in the past 100 years. Its run-up reached 6 m in the far field, and it wrought damage as far as the Chilean coast, and even in Antarctica, where it caused calving of a large iceberg (Brunt et al. 2011). A selection of the main points of scientific and societal interest includes:

  • Violation of Scaling Laws

    Detailed seismic tomography of the earthquake source, later confirmed by GPS studies and modeling of DART records, identified a patch of extremely large seismic slip on the fault, reaching \(\sim \)50 m over a length of no more than 100 km (Lay et al. 2011), thus clearly violating seismic scaling laws widely used in the assessment of tsunami risk.

  • Great Success in the Far Field

    Despite large run-up amplitudes leading to some structural damage, adequate warnings were given—and generally well heeded—in the far field, resulting to our knowledge in only two fatalities (one in Crescent City, California, and one close to Jayapura, Indonesia), which were preventable since both occurred in violation of mandatory evacuation orders. This positive outcome may have been helped by the daytime occurrence of the tsunami in Japan, which allowed immediate worldwide TV coverage of the catastrophic magnitude of the event (as opposed for example, to the night-time occurrence of the Chilean tsunami the previous year).

  • Fukushima Nuclear Accident

    The 2011 tsunami is of course known worldwide for the nuclear meltdown at the Fukushima Dai-Ichi nuclear plant, caused by cooling water starvation following the flooding of the emergency diesel power generators by the tsunami. As of 2019, and probably for decades to come, an exclusion zone of \(\sim 300\,\mathrm{km}^2\) is maintained, affecting about 50,000 residents.

    Careful analyses of the origin of the disaster by Nöggerath et al. (2011) and Synolakis and Kânoğlu (2015) have evidenced a cascade of design, scientific, engineering, and managerial blunders in the construction and operation of the plant, which suggests that the disaster was preventable, the best proof being that the Oganawa nuclear power plant, located only 127 km to the North, in a stretch of coastline which suffered considerably higher run-up, was safely shut down, and after a lengthy upgrade, is now scheduled to be re-started in the next few months.

  • Reasonable Success in Evacuation

    As offensive as this statement may sound in the wake of a 19,000 death toll, the evacuation process in the near field can be regarded as generally successful. It has been estimated that approximately 200,000 people were present in the areas totally eradicated by the tsunami, and thus that about 90% of them saved their lives by proper evacuation (Dengler, pers. comm., 2011; Fritz, pers. comm., 2011); this figure might even have been higher, but for horrific instances of orderly evacuation to shelters located at underestimated heights. It nevertheless constitutes a tribute to the awareness, education, and responsibility of the people of Japan, once again stresses the value of education in the mitigation of natural disasters, and stands in blunt contrast to the ineptitude underlying the whole Fukushima tragedy.

34. :

El Salvador, 27 August 2012

This tsunami went largely unnoticed as it resulted in very little damage and no fatalities, mostly because it occurred in an unpopulated segment of the Pacific coast of Central America. Yet, it ran up to 6.3 m, and most significantly, was identified as a tsunami earthquake through its deficient energy-to-moment (\(\Theta =-\,6.0\)) and energy-to-duration ratios (Borrero et al. 2014).

This is important, since it brings some support to the concept that the Central American coast, site of the 1992 Nicaragua event only 150 km away (see 1. above), could feature a regional preference for tsunami earthquakes. A rather obscure event on 26 February 1902 towards the Guatemalan border may also have shared some comparable properties (Cruz and Wyss 1983).

35. :

Haida Gwaii, British Columbia, 28 October 2012

This earthquake featured a rare shallow thrust fault mechanism at the Pacific-North America plate boundary, characterized in the area by extreme oblique convergence, and thus slip partitioning along the Queen Charlotte Fracture Zone (Lay et al. 2013). It featured locally substantial run-up values (up to 13 m), but took place in a seasonally unpopulated area (Leonard and Bednarski 2014). We dismiss the lone casualty, from a traffic accident in Hawaii caused by a drunken driver ramming into cars parked during the evacuation, as it is not a direct victim of the tsunami.

Like the 1996 Biak earthquake (see 9. above), this event underscores the possibility of significant tsunami hazard from rare thrust components to oblique subduction along partitioned boundaries featuring mostly strike-slip events.

36. :

Santa Cruz Islands, 06 February 2013

This event, which qualifies as a “tsunami earthquake”, triggered a tsunami running up to 11 m on Nendö Island, where it killed 10 people (Fritz et al. 2013). It is noteworthy that it shares its epicenter with another tsunami earthquake, on 21 July 1934, which led Okal and Saloor (2017) to identify systematic variations of coupling correlating with the slowness parameter \(\Theta \) along the Solomon-Vanuatu arc; however the 1934 event happened as an aftershock of a regular earthquake, 100 km to the South, while the 2013 event was primary.

37. :

Sea of Okhotsk, 24 May 2013

This event, the largest deep earthquake ever recorded (\(M_0 = 3.95 \times 10^{28}\)\({\hbox {dyn}}\,\times \,{\hbox {cm}}\)), did generate a tsunami recorded at millimetric amplitudes by two DART sensors off the Kuril Islands (Okal 2017). While it did not cause any damage, this unique detection raises the possibility of tsunamis observable at coastlines being generated by deep earthquakes, should even larger ones occur in the future.

38. :

Gwadar, Pakistan, 24 September 2013

This rather unusual scenario involved a tsunami in the Indian Ocean triggered by an earthquake occurring onland, about 200 km from the nearest shoreline. While of moderate size, the tsunami reached 1 m on the coast of Oman, and the process was accompanied by the emergence of a mud island off Gwadar, Pakistan. Heidarzadeh and Satake (2014) analyzed tidal gauge records of the event and concluded that the tsunami could not be caused by the emergence of the mud volcano, but rather by a probable underwater slump located about 70 km off the coast, at a total distance of \(\sim \) 400 km from the epicenter of the earthquake. This event serves as a reminder, if need be, of the capability of inland earthquakes to generate tsunamis through the triggering of underwater landslides; the 400-km distance is indeed large, but far from a record, since the 1910 Rukwa earthquake near Lake Tanganyika, in present day Tanzania, severed telegraphic cables in the Mozambique channel, at a distance of 900 km (Ambraseys 1991).

39. :

Iquique, Chile, 01 April 2014

This large event took place in what was defined as the Northern Chilean seismic gap (Chlieh et al. 2011), where the last major earthquake and tsunami took place in 1877. However, despite its large magnitude, it did not release the full accumulated strain, and thus seismic and hence tsunamigenic potential remains to this day (Hayes et al. 2014).

A remarkable aspect of the 2014 Chilean tsunami is that it apparently did not cause any casualties. This reflects the orderly, professional evacuation of about 1 million people from coastal areas in less than an hour, in stark contrast to the case of the Maule tsunami just 4 years earlier, and following a complete revamping of the country’s Tsunami Warning Center system.

40. :

Illapel, Chile, 16 September 2015

This new Chilean tsunami took place about 1000 km South of the 2014 epicenter. It produced much higher run-up reaching 13 m, and resulted in 8 fatalities (Aránguiz et al. 2016). As in the case of the 2014 event (see 39. above), a successful warning and evacuation took place, again involving about 1 million residents.

A remarkable experiment conducted during this tsunami was the in situ measurement of currents in the harbor of Ventura, California. Taking advantage of the \(\sim \)14-h travel time of the tsunami from Chile to California, Kalligeris et al. (2016) were able to deploy floaters equipped with GPS receivers, and documented current velocities of up to 1.5 m/s in turbulent flow. Because hazards to navigation in harbors depend crucially on current velocities, the collection of such data opens new opportunities for research critical to their further mitigation.

41. :

Taan Fjord, Icy Bay, Alaska, 18 October 2015

This remarkable event, fully described by Higman et al. (2018), shares many similarities with the classic 1958 Lituya Bay tsunami, 260 km to the SE, which splashed to a record height of 525 m (Miller 1960): it was generated by a massive landslide into a narrow fjord serving as estuary to an ice stream, and essentially faltered (because of extremely short wavelengths) upon reaching the open ocean. However, no detectable earthquake could be identified as its trigger, while the 1958 event followed a large shock (\(M_s =7.9\)). The field survey documented run-up reaching 193 m on the opposite side of the fjord. Seismic waves from the landslide were detected worldwide and inverted to suggest a slide mass of \(10^{11}\) kg.

42. :

Ecuador, 16 April 2016

This major earthquake occurred in a section of the South American subduction system characterized by diversity and complexity in the mechanism of strain release. The tsunami was only of marginal amplitude, reaching at most 10 cm on local tide gauges, and resulting in no fatalities. By contrast, the Colombia-Ecuador region was the site a mega-event in 1906, whose tsunami was recorded Pacific-wide and killed about 5000 people in the near field. It was followed by smaller shocks in 1942, 1958 and 1979, which released strain only over distinct fragments of the presumed 1906 rupture area. Only the 1979 event generated a substantial tsunami with about 600 victims. No mention can be found of a tsunami following the 1942 earthquake.

From the standpoint of its seismic source, and as detailed by Ye et al. (2016), the 2016 event may be a replica of the one in 1942, and the deceptive amplitude of their tsunamis explained by a relatively deep rupture, failing to reach the uppermost part of the subduction contact (Heidarzadeh et al. 2017). This brings the question of the nature of coupling at that shallow interface, which, if presently locked, could be ripening for a future event in the form of a tsunami earthquake. However, no such events are known in the historical record, which incidentally is very short in Southern Colombia and Northern Ecuador, with no events listed by Solov’ev and Go (1984) prior to 1906, despite Spanish presence going back to 1531.

43. :

\(Kaik\bar{o}ura\), New Zealand, 13 November 2016

This earthquake nucleated on land, but propagated along a complex series of offset faults, both on land and at sea, in the general transition zone between the Alpine Fault system and the Hikurangi Trough (Power et al. 2017). The tsunami was destructive both in the southern part of the rupture area, with a maximum run-up of 6.3 m, and farther South in the Banks Peninsula, as a result of the extreme indentation of the latter’s numerous bays. No comparable historical seismicity was known in the area, and thus the tsunami was unexpected. No casualties were attributed to the tsunami.

Modeling by Heidarzadeh and Satake (2017) has suggested a combined generation by elements of the complex dislocation and possible submarine landslides.

44. :

Karrat Fjord, Greenland, 17 June 2017

This tsunami was generated by the fall of an aerial landslide involving about 50 million \(\mathrm{m}^3\) into a fjord on the West coast of Greenland. At its source, the tsunami ran up to a height of 50 m across the fjord (Fritz et al. 2018). It attacked the village of Nuugaatsiaq, located 32 km away, running up 10 m and killing 4 persons. While no tidal gauges were available, a seismic station was being operated at Nuugaatsiaq, and its horizontal records were deconvolved by Paris et al. (2019) to reconstruct an equivalent tsunami amplitude time series in the channel facing the village. This documented a resonant oscillation of the channel comparable to observations in the Panama Canal by McNamara et al. (2011).

45. :

Gulf of Alaska, 23 January 2018

With a moment of only \(1.0\,\times \,10^{28}\)\({\hbox {dyn}}\,\times \,{\hbox {cm}}\), this intraplate earthquake was still the largest one recorded in 2018. It was characterized by an extremely complex rupture, involving several subsources located on offset segments of a primarily strike-slip fault structure (Lay et al. 2018), in the general vicinity of the slightly smaller events of 30 November 1987 and 06 March 1988. Like these two events, incidentally the first ones recorded by a prototype DART sensor (González et al. 1991), the 2018 earthquake generated only a mediocre tsunami, barely reaching 25 cm at Kodiak, and in the far field at Crescent City and in the Marquesas Islands.

46. :

Palu, Indonesia, 28 September 2018

With more than 2200 casualties, this tsunami stands as the most lethal one since the great 2011 Tohoku event. At the time of writing, its mechanism remains unclear. The geometry of the event advocates a “Mindoro-type” bayhead effect (see 7. above), where the strike-slip fault hits the coast of Sulawezi (Heidarzadeh et al. 2019); however visual evidence of underwater slides, in the form of surface eddies, suggests their contribution to the genesis of the tsunami (Harig et al. 2019).

47. :

Anak Krakatau, Indonesia, 22 December 2018

This catastrophic tsunami, killing upwards of 400 persons on the coasts of Java and Sumatra, resulted from the flank collapse of Anak Krakatau, as part of an eruption cycle which had started 6 months earlier. It is remarkable that the scenario had been forecast, at a quantitatively astonishing level of accuracy, by Giachetti et al. (2012). The flank collapse was well recorded seismically, and interpreted as a 0.3-\(\mathrm{km}^3\) slide, originating above sea level and propagating about 3 km under water (G. Ekström, pers. comm., 2019). In the absence of records from DART sensors (which did not trigger into higher sampling mode), horizontal seismograms at island stations can be deconvolved (Okal et al. 2019) to confirm amplitudes in the range of 0.1–1.5 mm on the high seas, due to the faltering, according to Green’s (1838) law, of a tsunami generated in extremely shallow waters.

Preliminary results from surveys suggest flow depths on the order of 2–5 m along the coast of Java, and run-up values of 30 m in the source area (Prasetya et al. 2019).

The 2018 tsunami differs fundamentally from the monstrous, worldwide one in 1883, which Press and Harkrider (1966) showed was due to an acoustic-gravity wave following an atmospheric explosion which they estimated at between 100 and 150 Megatons. In 2019, infrasound signals at the 10 Pa level were recorded at the Cocos Island infrasound array (1160 km away) from an explosion following the landslide by 7 minutes, but had already fallen below noise level at Diego Garcia, 3440 km away (Okal et al. 2019).

3 Discussion and Conclusion

In conclusion of this review, the main developments in the past 25 years can be summarized as follows:

  • We have acquired the capacity to obtain in quasi-real time (i.e., possibly before the tsunami reaches regional beaches) quantitative information on the earthquake source spectrum, both at low frequencies through W–phase inversions (Kanamori and Rivera 2008) and at high-frequency through the energy radiated by P waves, which can be compared either to seismic moment, or to source duration (Newman et al. 2011b).

  • We have developed algorithms to successfully model run-up as the non-linear interaction of tsunamis with originally dry land at the beaches, both in the near and far fields. As a result, countless scenarios have been prepared to assist Civil Defense authorities in the advanced mitigation of tsunami hazards (e.g., Barberopoulou et al. 2011), the resulting inundation maps providing a considerable improvement for the issuance of tsunami evacuation orders under operational conditions.

  • In the wake of the recent tsunami activity, and most particularly of the 2004 Sumatra event, the past 15 years have seen an explosion in the amount of high-quality data available under operational conditions to Tsunami Warning Centers, including broad-band seismic data and coastal and ocean-bottom sensors. We have developed real-time algorithms, such as SIFT (Gica et al. 2008), to successfully forecast tsunami amplitudes in the far field, introducing a quantitative component to the issuance of warnings for distant communities.

  • Possibly as a result of this progress, we have seen, since the 2004 Sumatra disaster, a spectacular reduction in the number of casualties in the far field (i.e., 1000 km or more from the epicenter), which amounts to a grand total of 2 in the past 14 years (not counting traffic fatalities during evacuation, which are not direct effects of the tsunami), both of which occurred in violation of mandatory evacuation. This also reflects a better education of the lay public worldwide to tsunami hazards, obviously the result of the catastrophic tsunamis of 2004 and 2011.

  • Elaborate techniques during post-tsunami surveys are starting to greatly improve our understanding of currents, i.e., particle flow velocities at beaches and harbors, which are the primary factors controlling momentum transfer, and therefore structural damage during inundation.

  • The examples of Montserrat and Krakatau show that accurate quantitative forecasts can be made even for complex sources such as flank collapses during episodes of volcanic activity.

However, there remain significant challenges which need to be addressed in the forthcoming years:

  • Paramount among them is the question of “tsunami earthquakes”, which have not been recognized effectively by warning centers in the past 15 years (e.g., Java, 2006; Mentawai, 2010; El Salvador, 2012). While an approach such as Newman et al.’s (2011b) is promising in this respect, a major challenge will be to instill its fundamental ingredient (the disparity between the duration and amplitude of shaking) in the minds of the populations at risk. In this context, it is remarkable that some survivors of the Mentawai disaster have reported to the Post-Tsunami Survey Team that the shaking had been “weak and long” (Hill et al. 2012). An emphasis on the ominous character of such a harbinger must be part of the tsunami education of coastal communities.

  • Finally, it is a bitter thought that the latest lethal tsunami took place during a catastrophic episode of a volcanic crisis which had taken months to build up, during which adequate seismic and geodetic monitoring could have been used to maintain awareness in the populations at risk. In hindsight, common sense would suggest that a closed building on an unprotected beach may not have been the best venue to hold a rock concert on the coast of Java, while Anak Krakatau was glowing red in the night sky, only 40 km away. Here again, a considerable effort of education is necessary in the communities at risk.