Detectability of seismic waves from the submarine landslide that caused the 1998 Papua New Guinea tsunami

On 17 July 1998, a tsunami caused serious damage on the northern coast of Papua New Guinea about 20 min after the mainshock of an Mw 7.0 earthquake. The tsunami has been attributed to a submarine landslide that occurred about 13 min after the mainshock because its arrival at the coast was too late and its height too great to be the direct result of the fault slip of the earthquake. Bathymetric data recorded after the tsunami revealed an amphitheater-like structure that was consistent with a recent submarine landslide. Most current tsunami warning systems are based on analysis of the early arrivals of seismic waves 5 generated by an earthquake. In this study we investigated whether evidence of the landslide could be identified in the coda waves recorded after the mainshock. Based on previous studies of the tsunami source, we constructed synthetic seismograms to represent the submarine landslide and compared them to the observed coda waves of the preceding earthquake, with particular attention to the period around 13 min after the mainshock. We found phases possibly corresponding to the landslide event. However, they were easily covered with coda waves from the mainshock. We concluded that the 1998 landslide was too small 10 to be evident in the coda waves following the magnitude 7 earthquake. Copyright statement. TEXT

3 Comparison of synthetic seismograms representing the 1998 submarine landslide with coda waves from the 1998 earthquake We estimated the amplitudes of the seismic waves generated by the 1998 landslide on the basis of the work of Watts et al.
(2003), as described below. Relatively large amplitudes were expected based on the model by Watts et al. (2003) because of its relatively large assumed mass.

5
When a landslide starts, the change in the force that has supported the landslide mass generates seismic waves. The force (F ) generated by a landslide is defined by F = ma, where m is the mass of the landslide and a is its acceleration. Watts et al.
(2003) assumed a characteristic time of 32 s. Tappin et al. (2008) estimated that the sliding process lasted about 100 s. We assumed that the sliding process was completed within several tens of seconds with an initial acceleration stage followed by 20 a deceleration stage (Fig. 4), and that the total impulse would not have been balanced because part of the deceleration stage would be affected by interaction of the sliding mass with sea water. The curves for each of the four intervals defined on Fig.   4 were expressed as trigonometric functions (Table 2). Five different time histories were considered ( Table 3). The peaks of ∫ F dt/m range from 3 to 11 m s −1 . The value 11 m s −1 of case (c) ( Table 3) is close to u max of Table 1. We constructed synthetic seismic records (Figs. 5 and 6) by applying the method of Takeo (1985) and using the seismic velocity model given 25 in Table 4.
The amplitudes of of our synthetic records are comparable to or larger than those of the seismic waves recorded after the mainshock of the M w 7.0 earthquake in the 50-100 s passband (Figs. 5 and 6). The phases indicated with bold lines could be the seismic waves from the landslide which caused the disastrous tsunamis. Whereas the ratios of the observed amplitudes to the synthetics at JAY were generally small, those at PMG were not so small. It is difficult to retrieve a consistent result about 30 the source parameters from this comparison. The amplitudes of the phase was smaller than the successive seismic wave at JAY. observed :::::::::: amplitudes :: of :::: coda :::::: waves :: at :::::: stations :: of ::::::::: epicentral ::::::: distance :::::: around ::::::: 2000km :::::: shown :: in :: 3 ::: for ::::::: passband ::::::: around :: 20 :: s. : It is unlikely to recognize the occurrence of the landslide based on the seismic records after the large earthquake.

Discussion
Several studies have shown that seismic waves generated by landslides can be detected. The landslide associated with the 1980 eruption of Mount St. Helens (Kanamori and Given , 1982;vol. 2  An extraordinarily large landslide mass would be required to prevent its signatures from being overwhelmed by the responses of the fault motion of a magnitude 7 or greater earthquake.
Other technologies with potential for direct detection of tsunami waves are tsunami radar (Barrick , 1979) and a fine barometer (Arai et al. , 2011), with which a tsunami is detected by reflection of electromagnetic waves or atmospheric pressure changes.

Conclusions
We investigated whether the tsunami-generating submarine landslide that occurred about 13 min after the 1998 PNG earthquake could be identified in the coda waves of the seismic data :: for ::: the ::::::: periods ::::: close :: to ::: the :::::::: landslide :::::::: duration. We constructed synthetic seismograms to represent the seismic signature of the landslide and compared them to the seismic data recorded after the earthquake, with particular attention to the period around 13 min after the earthquake. We found small seismic phases 15 possibly from the landslide. However, those phases were difficult to be recognized as an indication of a disastrous tsunami.
Other methods are needed to provide data for early warnings of tsunamis generated by submarine landslides of similar size (or smaller) to the one that generated the 1998 PNG tsunami. Networks of ocean-bottom tsunami gauges, similar to those provided by the DONET and S-net arrays in Japan, are among the likely candidates for this approach.
Data availability. Seismic data from the Jayapura seismic station are available at http://ohpdmc.eri.u-tokyo.ac.jp/. Seismic data from other 20 seismic stations shown in this study are available at https://www.iris.edu/hq/.
Author contributions. AKa analyzed the observed and synthetic seismic records and compiled most of the paper. YY suggested input to the methodology for construction of the synthetic records used in this study. KN researched previous studies of the 1998 PNG earthquake and tsunami. KF undertook preliminary research on the observed seismic records. MT, KT, TN, and AKo contributed to the research into previous studies of tsunamis caused by landslides and participated in related discussions.

25
Competing interests. The authors declare that they have no conflicts of interest directly relevant to the content of this article.
14 Table 2. Trigonometric source-time functions for each of the four intervals defined on Fig. 4.