Historical tsunami observability for Izu–Bonin–Mariana sources

The Izu–Bonin–Mariana Subduction System (IBM) is one of the longest subduction zones in the world with no instrumental history of shallow focus, great earthquakes (Mw > 8). Over the last 50 years, researchers have speculated on the reason for the absence of large magnitude, shallow seismicity on this plate interface, exploring factors from plate age to convergence rate. We approach the question from a different point of view: what if the IBM has hosted great earthquakes and no documentable evidence was left? To address the question of observability, we model expected tsunami wave heights from nine great earthquake scenarios on the IBM at selected locations around the Pacific Basin with an emphasis on locations having the possibility for a long, written record. Many circum-Pacific locations have extensive written records of tsunami run-up with some locations in Japan noting tsunami back to 684 CE. We find that most IBM source models should theoretically be observable at historically inhabited locations in the Pacific Basin. Surprisingly, however, some IBM source models for earthquakes with magnitudes as high as Mw 8.7 produce tsunami wave heights that would be essentially unobservable at most historically populated Pacific Basin locations. These scenarios aim to provide a constraint on the upper bound for earthquake magnitudes in the IBM over at least the past 400 years.


Introduction
Our understanding of tsunamis has grown over the past century following the instrumental observation of multiple, large, ocean-crossing tsunamis triggered by subduction zone earthquakes. Due to the destructive potential from violent shaking and the subsequent tsunami from these earthquakes, many studies have been conducted in the past 50 years on subduction zones globally and their ability to produce large earthquakes, as not all subduction zones are thought to be capable (Singh et al. 1981;Goldfinger et al. 1992;Johnson and Satake 1999;Barnes et al. 2002;Satake and Atwater 2007). Although initially proposed as an explanation for the presence or absence of back-arc spreading, the classification of subduction zones into Mariana-type and Chilean-type appeared to provide further explanatory power with regard to the mode of seismic strain release at each boundary (Uyeda and Kanamori 1979). Subsequent studies continued to develop the Chilean-Mariana subduction style classification paradigm (Ruff and Kanamori 1980;Kanamori 1986;Ruff 1989;Stern 2002;Stern et al. 2003). In particular, regression analysis of slab age, convergence rate, and maximum earthquake magnitude all seemed to confirm this classification scheme (Ruff and Kanamori 1980). One of the outcomes of this regression analysis was the prediction that some subduction zones, including northern Japan, Sumatra, and Cascadia, had a maximum earthquake magnitude less than Mw 9 (Heaton and Kanamori 1984;Kanamori 1986). In the case of the Mariana and Izu-Bonin subduction systems, the predicted maximum earthquake magnitude according to the Chilean-Mariana subduction style classification paradigm is even smaller, possibly less than Mw 7.6 (Kanamori 1986).
The realization that the Cascadia subduction zone had experienced a great earthquake in 1700 (Satake et al. 1996) along with the subsequent great earthquakes in Sumatra (2004) and northern Japan (2011) provided direct refutations of the predictions from this classification scheme. Following these earthquakes, numerous researchers began to question the assumption that some subduction zones cannot host great earthquakes (McCaffrey 2008;Stein and Okal 2011;Kagan and Jackson 2013;Rong et al. 2014). Under this new wave of thinking that all subduction zones can host large tsunamigenic earthquakes, the physical dimensions of the subduction zone itself are thought to be the best predictors of maximum earthquake magnitude. For the Izu-Bonin-Mariana subduction system, this approach yields a value of Mw 9.2 for the Izu segment and Mw 9.5 for the Mariana segment (McCaffrey 2008). McCaffrey (2008) also provides a prediction of recurrence interval with estimates of 723 years for the Izu segment and 893 years for the Mariana segment.
Given the limited time span of the instrumental seismic record, extending our knowledge of great earthquakes further into the past must rely on historical documentation of seismic events, mostly in the form of felt reports of shaking and inundation from tsunami. This approach has yielded immense knowledge in regions where the written record is long, such as Japan (Ishibashi 2004;Matsu'ura 2017). However, for great earthquakes in remote or sparsely populated regions, the historical record is likely incomplete. This study addresses the possibility of identifying historical earthquakes in the Izu-Bonin-Mariana region from the standpoint of tsunami visibility in the far-field: if a large earthquake occurred in the Izu-Bonin-Mariana region, which areas of the Pacific Basin would be most likely to have recorded the effects of the ensuing tsunami? We then interrogate the historical record for some of these locations looking for signs of recorded tsunami.

Background
The instrumental earthquake history of the IBM is limited (< 130 years) with the largest recorded earthquake being a Mw 7.8 on August 8, 1993, with a focal depth of nearly 60 km ( Fig. 1) (US Geological Survey 2017; Harada and Ishibashi 2008). This earthquake, along with earthquakes of Mw 7.0 and 7.1 in 2001 and 2002 near Guam have been the subject of interest as being some of the largest magnitude shallow forearc seismicity in the Mariana region (Tanioka et al. 1995;Campos et al. 1996;Harada and Ishibashi 2008). Farther north, along the Izu-Bonin segment of the IBM, the largest instrumentally recorded earthquake was the December 21, 2010 Mw 7.4 Bonin Islands earthquake near Hahajima, Japan (US Geological Survey 2017). Although this earthquake was a shallow-focus outer-rise normal faulting event in the Pacific Plate, and not on the subduction interface, it nonetheless caused a 7-13 cm tsunami along the south coast of Honshu (National Tsunami Warning Center).
The tsunami history for the Izu-Bonin and Mariana Islands, including the Caroline Islands and Guam extends back to at least 1606 with the description of a tsunami accompanying a volcanic eruption on Hachijō-jima in the northern Izu Islands (Fig. 1) (Soloviev and Go 1984a, b). The earliest tsunami attributable to an earthquake from the Izu-Bonin-Mariana arc is likely the earthquake and tsunami of January 1826, whose shaking and tsunami effects were felt strongly on Chichijima in the Bonin Islands ( Fig. 1) (Soloviev and Go 1984a, b). A larger and more destructive earthquake, with a long-lived aftershock sequence and tsunami, occurred near Guam in January of 1849 (Fig. 1). The mainshock of this earthquake razed all of the masonry structures on the island and caused a tsunami that may have inundated Guam for a distance of 500 m and may also have been recorded on the atoll of Satawal in the Caroline Islands over 700 km SSE of Guam ( Fig. 1) (Soloviev and Go 1984a, b;Lander et al. 2002). Additional earthquakes and tsunami were recorded in the Bonin Islands on the island of Chichijima in the fall of Location of the Izu-Bonin-Mariana subduction system. Stars represent the four largest, shallow focus, instrumentally located earthquakes in the subduction system. Circles represent locations mentioned in this manuscript. Colored rectangles denote faults from the NOAA SIFT fault model database. The colors represent the fault segments used in this paper: the green segment is the Izu-Bonin segment, the purple segment is the north Mariana, and the orange is the south Mariana segment 1872 (supposedly also recorded in the Hawaiian Islands, although this is less certain. See Cox and Lander (1994) and references therein for further discussion) (Cholmondeley 1915;Soloviev and Go 1984a, b), and 1892 (felt in Guam) (Soloviev and Go 1984a, b). Subsequent tsunamis in the region appear to have been volcanic or meteorologic in origin (Soloviev and Go 1984a, b). More recently, the instrumentally recorded Mw 7.8 earthquake of August 8, 1993 produced a tsunami with wave heights of around 20 cm on a tide gauge at Mera, Japan, on the southern end of the Bōsō Peninsula (Tanioka et al. 1995). Local effects from the August 8, 1993 tsunami were more pronounced on the island of Guam (Sigrist 1995).

Modeling
To assess far-field tsunami visibility, we produce generalized models of tsunami generated by large, shallow subduction zone earthquakes from the IBM. Using the IBM subduction zone fault geometry from the National Oceanic and Atmospheric Administration (NOAA) Shortterm Inundation Forecasting for Tsunami (SIFT) fault model database (Gica et al. 2008), we model earthquakes with magnitudes from Mw 8.7 to Mw 9.3 with smoothly varying slip centered at different locations along the IBM. We then use the resulting surface deformation as a starting model for GeoClaw V5.7 to model the resulting tsunami waveform across the Pacific Basin. Time series of water surface height are then calculated at select, shallow water locations around the Pacific Basin as a proxy for tsunami arrival height.
In order to characterize the general tsunami behavior of earthquakes in the IBM, we begin by dividing the IBM into fault segments. Since the majority of a tsunami's energy is directed perpendicular to the strike of its causal earthquake (Ben-Menahem and Rosenmen 1972), we partition the IBM into segments based on the overall strike of the subduction system. This partitioning allows us to study the overarching directivity patterns of IBM-generated tsunami but is not intended to imply the existence of barriers to slip or other rupture kinematic behavior. Ruptures that span the fault segments highlighted here should produce tsunami with directivity patterns that display characteristics of each segment in a manner similar to what is seen with the NOAA SIFT approach (Gica et al. 2008). Following the large-scale strike of the IBM, we partition the subduction zone into three segments ( Fig. 1): From north to south, an Izu-Bonin segment (Fig. 1, green fault); a northern Mariana segment (Fig. 1, purple fault), and a southern Mariana segment (Fig. 1, orange fault).
For each of these fault segments, we then model three slip distributions with magnitudes Mw 8.7, 9.0 and 9.3, each with a smooth Gaussian-like slip profile along strike (Additional file 1: Table S1 and Additional file 2: Figure  S1). For each dislocation source and for each fault segment, we simulate the propagation of a tsunami to each circum-Pacific location listed in Table 1. In choosing circum-Pacific locations for forward modeling, we endeavored to select locations with known written records of past tsunami. Based on these criteria, 23 locations were selected (Fig. 2, Table 1). Tsunami simulations were then performed using GeoClaw 5.7 (Clawpack Development Team 2020) with grid refinement guided by the adjoint method (Davis and LaVeque 2016). Bathymetric grid refinement cell sizes ranged from 1º to 30 arcsec using bathymetry from GEBCO 2019 (GEBCO Compilation Group 2019). Example tsunami waveforms are shown in Additional file 3: Figures S2 and Additional file 4: Figure  S3.
Since GeoClaw solves the depth-averaged shallow water equations, they provide the best approximation of water height when the wavelength is long relative to the depth of the water. In order to capture wave heights in shallow water while retaining the full waveform, we attempted to keep our gauge locations in water depths shallower than 20 m. Additionally, GeoClaw does not estimate tsunami dispersion. For our tsunami scenarios, the rupture widths are approximately 90 km and lie in water that is, on average 4000 m deep. Using these values in Eq. (1) of Shuto (1991), we expect tsunami dispersion to be negligible to distances of at least 4000 km. Beyond this distance, preliminary modeling using JAGURS (Baba et al. 2015) suggests that dispersion may result in a limited reduction in amplitude in dispersed wave trains at the furthest distances from the tsunami source (Additional file 5: Figure S4). Similar results can be seen for trans-oceanic paths for the 2011 Tohoku tsunami (Baba et al. 2017). Thus, maximum wave height model results using output from GeoClaw will represent a worst-case scenario at the furthest gauge locations.
Finally, in order to assess the far-field impact of each tsunami scenario, we have developed a basic tsunami observability index based on the relative amplitude of the largest wave with respect to the local tidal variation. Since we are interested in historical observability, the mean range of tides for a location, defined as the difference in height between the mean high water and the mean low water (Gill and Schultz 2001), provides a reasonable benchmark that an observer familiar with the typical tidal variations at a location might use in recognizing an incoming tsunami. Using tidal datum information provided by the GLOSS Network (Caldwell et al. 2015), the mean range of tides was identified for each location and is listed in Table 1.
Our tsunami observability index consists of three observability categories, Unobservable, Possibly   Figure 3 shows the results of the modeling for each of the faulting scenarios. Earthquakes with magnitude Mw 9.3 produce tsunamis with wave heights in excess of the local tidal range at nearly all Pacific Basin locations and therefore fall into the category of Likely Observable. For Mw 9.0 earthquake scenarios, nearly all locations in Japan show wave heights in excess of twice the tidal range (Table 1, Fig. 3), suggesting that observers in Japan would have had the opportunity to observe tsunami from these earthquakes. Similarly, for observers on the west coast of Filled circles represent tsunami visibility with the darkest purple indicating maximum wave heights in excess of the local tidal range, medium purple representing maximum wave heights larger than 30% of the local tidal range, and the lightest purple indicating maximum wave heights less than 30% of the local tidal range. All plots use the same color scale for maximum wave height the America's, nearly all locations show maximum wave heights in excess of the local tidal range for sources in the Izu-Bonin and North Mariana segments, while sources from the South Mariana exceed the local tidal range only at lower latitudes. For Mw 8.7 earthquake scenarios, Izu-Bonin sources direct tsunami wave energy towards locations near Shikoku, where maximum wave heights rise well above the local tidal range, however locations elsewhere in Japan experience maximum wave heights below the local tidal range (Table 1). Along the west coast of the Americas, Izu-Bonin sources also produce smaller wave heights, with most locations at around 50% of the tidal range. In contrast, Northern Mariana sources produce large wave heights at Crescent City, CA, while wave heights in Japan exceed the tidal range at Tanabe (Wakayama Prefecture) and along the northeast coast of Honshu. For Northern Mariana sources wave heights elsewhere in Japan are less than the tidal range. Finally, Southern Mariana sources produce mostly Possibly Observable wave heights in Japan with a localized large maximum wave height in Tokyo Bay (Futtsu, Chiba Prefecture) likely due to topographic wave guiding by bathymetric structures in the Izu-Bonin back-arc (Satake and Kanamori 1991).

Hypothetical observability of large Mariana tsunamis
For many locations, the Mw 8.7 sources produce maximum wave heights of about half of the local mean range of tides. At this amplitude, it is uncertain whether tsunami would be consistently observed at all hours of the day, under all tidal stages, or in all weather conditions. Table 1 lists our present knowledge of the earliest written record of tsunami for each location we have considered. The presence of written records at these times suggests that records of notable natural phenomena were kept in each region and that these records exist to the present day. In seeking out the locations with the longest written records of such events, we are attempting to place bounds on the length of time of quiescence of the Izu-Bonin-Mariana system. We do note that, although many locations have written records of tsunami dating back hundreds of years, this does not indicate that these written records are complete. The completeness of any written sources depends strongly on both preservation and access, including physical and linguistic access (e.g., Ambraseys and Melville 1982; Martin et al. 2020). The transmission and survival of historical tsunami data depends on historical and geographical circumstances which are not consistent throughout a region or across time. For example, examination of Fig. 3e-i suggests that the north coast of New Guinea would experience wave heights in excess of 2 m for both North and South Mariana sources larger than Mw 9.0 and possibly in excess of 5 m for Mw 9.3 sources. However, these locations were not modeled with gauges due to the presumed lack of a written historical record for northern New Guinea prior to the instrumental period (Lutton 1991;Nagle 2002). Similarly, for Northern Mariana tsunami sources, locations along the east coast of the Philippines would experience wave heights in excess of 5 m. However, in contrast to New Guinea, the Philippines possess a large, though untapped archive of Spanish documents, although pre-Spanish written history records appear to be nonexistent (Punzalan 2006). Finally, many cultures in the Pacific Basin retain legends and folklore surrounding earthquakes and tsunami; the full appreciation of which is only starting to be realized (e.g., Ludwin et al. 2007). Given all these caveats, we note that the present study is not intended to suggest that the lack of a tsunami record is evidence that no tsunami occurred (an argument from ignorance), rather, our focus is on identifying locations that should prove to be fruitful areas for future historical research.

Observability compared to the historical record
The earliest written records in the Pacific Basin we considered pertain to the earthquake and tsunami of 684, which occurred in the Kochi prefecture of Japan (National Geophysical Data Center 2022). Tsunami from this earthquake was recorded at Tosa, Japan, which is represented by our gauge location in Urado Bay near present day Kochi (Table 1, Gauge 6). Records in the Kochi Prefecture subsequently show tsunami in 1605 and then in 1707. To the east, in Wakayama Prefecture, Japan, records for a tsunami in 869 exist at Iwashiro, near our Tanabe Gauge (Table 1, Gauge 5). The 869 tsunami at Iwashiro was followed by a tsunami in 1099. The number of records increases starting around 1403 indicating an improvement in the completeness of the observations for this location. Further east in Shizuoka Prefecture, Japan, the earliest record is of a tsunami in 1096 with subsequent records increasing in frequency beginning in 1498. Tsunami in Shizuoka Prefecture is represented by our Miho gauge (Table 1, Gauge 7). Near the entrance to Tokyo Bay, in Chiba Prefecture, Japan, the earliest recorded tsunami is from 818. However, consistent records of tsunami do not begin at that location until 1602. Northward, along the Pacific Coast in Iwate Prefecture, Japan, records of the 869 tsunami exist, followed by tsunami in 1088 and 1257. In this region along the northwest coast of Honshu, the record of tsunami appears to be more complete beginning around 1611. Figure 4 summarizes the historical record of tsunami at our Japanese gauge locations.
Along the east coast of mainland China, the earliest recorded tsunami in Shanghai is in 1498 (Zhao et al. 2017;National Geophysical Data Center 2022). Followed by tsunami in 1668, 1854, 1867, and 1983(Zhao et al. 2017; National Geophysical Data Center 2022).
Along the west coast of the Americas, the earliest recorded tsunami is the 1562 tsunami observed at Concepción, Chile. Tsunamis in the BioBío region around Concepción are then regularly recorded up to the present day at a rate of 2.5 events per century. Further north along the South American coast, the earliest recorded tsunami in Lima, Peru is in 1586. Similar to records at Concepción, eyewitness reports of tsunami at Lima continue to the present day, averaging 2.5 events per century.
The earliest written tsunami records for other locations around the Pacific Basin are more recent, with records in Manila, Philippines, in 1645;Acapulco, Mexico, in 1732;Keelung, Taiwan, in 1754;Honolulu, Hawaii, in 1812; and Auckland, New Zealand (Henderson Bay), in 1835 (Table 1).
By taking into consideration the length and completeness of historical tsunami observations along with the predicted wave heights, we come to the following conclusions. For Izu-Bonin tsunami sources, the most reliable locations for observation are Tanabe (Wakayama Prefecture) and Kochi (Kochi Prefecture). These locations show large wave heights relative to the local tidal range for earthquake sources down to Mw 8.7. Further, tsunami from Mw 9.0 and larger earthquakes in the Izu-Bonin segment are predicted to be visible at all locations throughout Japan, as well as along the West Coast of the  Americas. It is also notable that Hilo, Hawaii shows consistently large wave heights for magnitudes down to Mw 8.7 as well, particularly in comparison to wave heights at Honolulu, Hawaii (Additional file 3: Figure S2). Thus, given the apparent completeness of records in Wakayama Prefecture back to 1403, we postulate that records of any tsunami from an earthquake greater than Mw 8.7 from an Izu-Bonin source after 1403 should be present in the historic record.
For North Mariana tsunami sources, the most reliable locations for observation are Tanabe (Wakayama Prefecture) and Kuwagasaki (Iwate Prefecture), both of which show a high likelihood of observation down to Mw 8.7. Tsunami from Mw 9.0 and larger earthquakes are predicted to be visible at all locations throughout Japan, as well as at most locations along the West Coast of the Americas. Similar to Izu-Bonin sources, wave heights at Hilo, Hawaii are also consistently large. Given the apparent completeness of records in Wakayama Prefecture and Iwate Prefecture back to 1611, we postulate that any tsunami from an earthquake greater than Mw 8.7 from a North Mariana source possibly after 1403 and certainly after 1611 should have been recorded in the historic record.
For South Mariana tsunami sources, the most reliable locations for observation are Futtsu (Chiba Prefecture) and Otsuchi (Iwate Prefecture) both of which show a high likelihood of observation down to Mw 8.7. Tsunami from Mw 9.0 and larger earthquakes are predicted to be visible at all locations throughout Japan and at low latitude locations along the West Coast of the Americas. Given the apparent completeness of records in Chiba Prefecture and Iwate Prefecture back to 1241, we postulate that any tsunami from an earthquake greater than Mw 8.7 from a South Mariana source possibly after 1241 and certainly after 1611 should have been recorded in the historic record.
For locations in the Americas, our modeling suggests that Lima, Peru, is the most susceptible to large waves from all sources for earthquakes larger than Mw 9.0. Further south, Concepción, Chile appears to be susceptible to large waves from Izu-Bonin sources larger than Mw 9.0 and from Northern and Southern Mariana sources larger than Mw 9.3. Thus, we postulate that any tsunami from an earthquake greater than Mw 9.3 for all sources after 1562 and, for select sources greater than Mw 9.0 after 1586 should have been recorded in the historic record.

Candidate IBM earthquakes in the historical record
Interrogating the historical record reveals at least two candidate great subduction zone earthquakes with a source in the IBM: the 1605 Keicho earthquake, and the 1849 Guam earthquake. Considerable attention has been given to the 1605 Keicho earthquake, in part because its effects were experienced along the Nankai region of Japan, a region with numerous great earthquakes in the historical era (Seno 2002;Ishibashi 2004;Ando and Nakamura 2013;Fujino et al. 2018). Out of all of the Nankai Trough earthquakes and tsunami, the 1605 Keicho earthquake is notable for its lack of strong seismic shaking compared with the size of its tsunami spurring interest in whether the 1605 earthquake was a tsunami earthquake (Seno, 2002;Ishibashi, 2004). Following reports of shaking felt in the Kanto and Tohoku districts along with anomalously large tsunami from the 21 December, 2010 Bonin Islands earthquake (Fig. 1)-a shaking and tsunami pattern similar to the 1605 earthquake- Harada et al. (2013) suggested that the 1605 earthquake may have had a source in the Izu-Bonin Islands, rather than the Nankai Trough. Based on the focal mechanism of the 2010 Bonin Islands earthquake, Harada et al. (2013) use a steeply dipping outer-rise fault to model tsunami wave heights to the Nankai region and found good agreement with tsunami reports for the 1605 earthquake. Our modeling of great earthquakes from the Izu-Bonin segment of the IBM is in agreement with that of Harada et al. (2013), showing large tsunami along the south coast of Honshu for sources down to Mw 8.7. Our models suggest that earthquakes larger than Mw 9.0 from the Izu-Bonin segment are expected to be observable throughout Japan, including the northeast coast of Honshu. Since there is no indication that the 1605 tsunami was recorded outside of southern Honshu, it is unlikely to have been as large as Mw 9.0. Finally, although it may be difficult to discriminate between tsunami generated by outer-rise normal faulting earthquake and shallow thrust earthquakes in the historical record, additional records of the 1605 tsunami may shed more light on the source region.
The second candidate great earthquake from the IBM system is the 1849 Guam earthquake. As mentioned previously in the Background Section, the 1849 Guam earthquake destroyed all of the stone structures on the island, created sand boils that discharged sea water, and caused a tsunami that inundated the island in places, up to 402 m (Soloviev and Go 1984a, b;Lander et al. 2002). Both the earthquake and tsunami were also observed on the island of Satawal, over 700 km SSE of Guam, where locals claim to have survived a great earthquake and ensuing flood (Lander et al. 2002). These observations suggest that an ocean-crossing tsunami was generated by this earthquake and should have been observable throughout the Pacific Basin. Our modeling suggests that earthquakes from the South Mariana segment with magnitude larger

Sensitive locations
Some locations seem especially sensitive to IBM earthquakes with magnitudes as low as Mw 8.7 producing wave heights in the Likely Observable category. For example, modeling predicts wave heights in Hilo, Hi of 7.16 m from an Mw 9.3 Izu-Bonin earthquake, at an astonishing 1400% of the local tidal range (Table 2). Other locations, such as Toucheng, Taiwan, on the northeast coast of Taiwan could experience wave heights of 4 m for Mw 9.3 Izu-Bonin sources (835% of the tidal range). The wave heights in Toucheng, Taiwan, stand in contrast to those of nearby Keelung, Taiwan, a major port which appears to gain some protection from IBM tsunami by virtue of its location in the lee of the Yilan peninsula (Additional file 3: Figure S2). Both the wave heights at Hilo and the Toucheng appear to be due to focusing of wave energy by undersea bathymetry. This effect has been noted by others at Hilo, Hawaii (e.g., Tang et al. 2010).
Other locations, such as Tanabe, Japan are predicted to experience Likely Observable wave heights for all faulting scenarios at all magnitude except South Mariana sources with magnitudes of Mw 8.7. Otsuchi, Japan, is similar in that it is expected to experience Likely Observable wave heights for all sources and magnitude except Izu-Bonin sources with magnitudes of Mw 8.7. In both locations, the largest wave arrival varies between scenarios suggesting a complex interplay of bathymetric, wave reflection, and resonance effects (Additional file 4: Figure S3). The consistency of these observations suggests that paleoseismic studies in these areas should consider IBM sources.

Other measures of observability
Our approach to estimating tsunami observability focuses heavily on comparing the predicted tsunami wave heights to the local tidal range. However, coastal residents can be in tune to other changes to the ocean that would also indicate the arrival of a tsunami.
For example, many tsunami reports in the NCEI/ WDS database note changes in the speed and direction of the currents within harbors. Tsunami modeling software, such as GeoClaw, can calculate the horizontal flow velocity, and future research could explore the interplay between tsunami observability and horizontal flow velocity at low wave-height locations.
Another factor influencing the eyewitness observation of tsunami is the time of arrival. For tsunami arriving at night, the first (and often largest) waves can be missed. This appears to be the case with the tsunami from the 1906 Ecuador earthquake, which was observed in Gisborne approximately 8-10 h later than expected leading some to suggest that the initial, and largest, waves arrived during the night and went unnoticed (GNS Science 2020). Similarly, the combination of a nighttime arrival and an arrival during low tide for the first waves from the tsunami generated by the 1868 Peru earthquake as observed at various locations throughout the South Island of New Zealand led to poor observation of the wave heights and arrival times there (GNS Science 2020).
The ability to observe tsunami can also be greatly affected by how calm the sea is during the arrival of the waves. For example, the 28 Feb 1973 Kuril Island tsunami produced only fragmentary eyewitness reports of tsunami run-up mostly due to stormy weather in the region (National Geophysical Data Center 2022). In the historical record, stormy seas can both hinder observation of tsunami and create the false impression of a tsunami (e.g., a meteotsunami).

Conclusions
We have modeled tsunami from megathrust sources in the Izu-Bonin and Mariana subduction zone and calculated maximum wave heights for select locations Water heights are shown as a percentage of the local mean range of tides as determined from nearby sea level monitoring stations from the GLOSS network (see Table 1) around the Pacific Basin with a focus on those with long written histories of tsunami observations. For an earthquake larger than Mw 9.0, the resulting tsunami would be visible at a large number of locations in the Pacific Basin. Assuming the completeness of written records of tsunami at our chosen locations, this would constrain the magnitude of the largest earthquake from the Izu-Bonin-Mariana system to less than Mw 8.7 since at least 1605. For earthquakes smaller than Mw 8.7 few farfield locations in the Pacific Basin would be expected to receive maximum wave heights sufficient to be observable without marigrams. Future research into written historical records of tsunami inundation around the Pacific Basin should therefore consider IBM tsunami sources with Mw > 9.0.