1 Introduction

Eastern Indonesia, and the Banda Sea in particular, is a region of very active and complex tectonics (Hamilton 1979; McCaffrey 1988; Spakman and Hall 2010; Pownall et al. 2013). Despite a historical record rich in major, destructive earthquakes and tsunamis, during the more recent era of instrumental seismology most of the major events have occurred in western Indonesia. The only way to better understand the tsunami threat in eastern Indonesia is therefore to glean as much information as we can from the historical record, which often consists of accounts that are sparse and difficult to interpret.

The oldest detailed tsunami account in Indonesia was documented by Rumphius (1675). A devastating earthquake rocked Ambon and its surrounding islands on 17 February 1674. The earthquake was followed by a massive tsunami about 100 m in run-up height which was only observed on the northern coast of Ambon Island while other areas experienced only minor tsunamis. The earthquake and tsunami caused more than 2300 fatalities, mostly on the northern shore of Ambon.

The source of the tsunami and earthquake is unknown. Løvholt et al. (2012) and Harris and Major (2017) speculated that it was triggered by an earthquake from south of Ambon and a landslide triggered by an earthquake from inside Ambon Bay, respectively. However, no attempt has been made to investigate this event further, particularly to answer why the extreme run-up was observed only on the northern coast of Ambon. Therefore, the primary source of the tsunami and earthquake remains open to question.

In the following sections of this paper, the tectonic setting around Ambon is discussed first, followed by our interpretation of the accounts of this event. The primary source of the earthquake and tsunami is investigated through analysis of the Rumphius document. Tsunami modelling is then performed to confirm the analysis. Lastly, the result of the analysis and the implications of the findings are discussed.

2 Tectonic Setting Around Ambon

Ambon is a small volcanic island that lies southwest of Seram Island (Fig. 1). It consists of two small islands, Hitu and Leitimor, which are connected by a short isthmus (Fig. 2). Ambon is located to the southwest of Seram Island which is part of the outer Banda Arc. The islands are surrounded by major faults, namely the Seram Megathrust, Kawa Fault and the Banda Detachment (Fig. 1).

As summarised in Patria and Hall (2017), the Seram Megathrust stretches from Kai Islands in the east side to the northwest of Seram Island (Fig. 1). It is often described as a subduction zone (e.g. Hamilton 1979; Honthaas et al. 1998). However, others have argued that it is a foredeep produced by loading from a developing fold and thrust belt (Audley-Charles et al. 1979; Pairault et al. 2003; Spakman and Hall 2010). Through high-resolution bathymetry and seismic data, Patria and Hall (2017) confirmed the second hypothesis: that it is a result of oblique intra-plate convergence.

The Kawa Fault is a prominent structure in central Seram Island (Fig. 1). The fault runs from Piru Bay, off the northern shore of Ambon, to the central south of the island on the northern side of the Banda Sea. The fault has a major left-lateral movement identified through geological observations (Pownall et al. 2013). A large earthquake which caused a catastrophic tsunami on Seram Island in 1899 was suspected to have ruptured the Kawa Fault (Soloviev and Go 1974).

At the southern end of Seram, Pownall et al. (2016) argue that the Kawa Fault continues to the Banda Sea and is related to the Banda Detachment (Fig. 1). The Banda Detachment is a recently discovered low-angle normal fault identified from geological observations and high-resolution bathymetry data. Moreover, the Banda Detachment exhibits rapid extension as indicated by a very thin sediment layer in the Weber Deep (Pownall et al. 2016).

On Ambon, Watkinson and Hall (2017) identified several faults via a digital elevation model of the Shuttle Radar Topography Mission (SRTM). First, a normal fault with a very steep angle was identified on the northern shore of the island. Then a lineament with NE–SW trend, crossing the city of Ambon was observed. Last, a suspected quaternary normal fault was observed on the southern side of Ambon Bay.

Fig. 1
figure 1

Tectonic setting of Ambon and its surrounding islands; SMT Seram Megathurst, KF Kawa fault, SST purported South Seram Thrust, BD Banda Detachment, WT Wetar thrust, WAF Wetar-Atauro fault

Full size image

Ambon and its surrounding islands (Seram, Haruku, Saparua, Nusa Laut, and Buru) have experienced at least 14 tsunamis in the historical past (Table 1). There were two events each generated by volcano activities and landslides, earthquakes (seven events), and three events with an undetermined source. The majority of the events are categorised to be minor tsunamis with a maximum height up to 2 m, except for the 1674 and 1899 events. The 1674 tsunami event was started by strong ground motion felt on Ambon Island. An extreme tsunami run-up up to 100 m was observed only on the southern shore of the island with minor tsunamis in other areas (Rumphius 1675). The source of this tsunami will be discussed in the following sections. In the 1899 event, settlements on Seram Island were devastated by an earthquake that was suspected of originating from the Kawa Fault (Soloviev and Go 1974). It triggered multiple landslides which generated tsunamis along the northern shore of Seram Island. The tsunami rose up to 9 m at villages of Amahai, Paulohi, and Elpaputih, with no reports of tsunami on the north coast of Ambon.

Table 1 Historical tsunamis on Ambon and its surrounding islands between 1659 and 1950
Full size table

3 Historical Accounts of the Ambon Island 1674 Earthquake and Tsunami

Historical accounts of the 17 February 1674 Ambon earthquake and tsunami were documented in the book “Waerachtigh Verhael Van de Schlickelijcke Aerdbebinge” written by Rumphius (1675). The book was translated into English with the title “The true history of the terrible earthquake” (Rumphius 1997). The historical accounts described below are based on the English translation and summarised in Fig. 2 and Table 2.

Fig. 2
figure 2

Historical accounts of the 1674 Ambon Island a earthquake and b tsunami. Location of map b is shown as the red dashed box. Colour circles represent interpreted earthquake intensity. Tsunami heights are indicated by the gold (100 m) and blue (5 m) color bars

Full size image

The event occurred at about 7:30 p.m. local time when people in Laitimor were celebrating Chinese Lunar New Year. In Ambon, the bells in Victoria Castle swung by themselves and people who were standing fell to the ground as the earth heaved up and down like the sea. Stone buildings collapsed and buried up to 80 people. Strong shaking was felt on the mountains in Laitimor, with rocks falling and ground cracking open. In Hutumuri near the coast, on the eastern side of Laitimor, seawater burst into the air like a fountain.

The earthquake was reported from Hitu as well. In Waytome River, on the northern side, the river water spurted to 6 m high. People in the north to northwest of Hitu heard a loud sound like canon-fire. They noticed two long, thin marks in the sky, extending from Luhu to Seith, shortly before the earthquake. Less than 15 min after the earthquake, villages between Lima and Hila were wiped away by a gigantic mountain of seawater. The seawater rose about 50 to 60 fathoms (approximately 90–110 m) to the top of the surrounding hills, and more than 2300 people perished.

This unusual phenomenon was reported at other places, but with much less intensity. Hitu Lama village, located approximately 15 km to the east of Hila, reported that the seawater rose only about 3 to 5 m, killing 35 people. A little further to the east, 40 houses in Mamala were swept away without fatalities. The settlement of Orien (present name; Ureng), which is located < 10 km west of Lima reported that the seawater rose and inundated land but it did not enter the houses. People in Larike, a village in the westernmost of Hitu, observed that the seawater rose < 1 m at the Rotterdam Redoubt. In the southern and eastern areas of Ambon Island, there was much less seawater oscillation reported, apart from some small boats being tossed over each other.

At Luhu in Seram Kecil, the seawater inundated trees and the dwellings of a company. The water rose to a height just over 5 m. At the northernmost tip of Piru Bay, half of the houses in Tanuno were engulfed by water, but without fatalities. Fishermen in Piru Bay said the sea remained calm with a noticeable ripple. A much lower seawater oscillation was noticed by people from Manipa, Salati, Haruku, Nusa Laut, and Banda Neira Island compared with the oscillation that was observed in Hitu and Laitimor. Aftershocks continued for at least 3 months. The two largest aftershocks occurred on 6 and 10 May.

What were the source of the ground motion and seawater phenomena described above, particularly on 17 February 1674? In the following sections, we will make use of the historical accounts of ground motion and tsunami observation to answer that question.

Table 2 Tsunami and ground motion interpretation from Rumphius (1997)
Full size table

4 Source Identification

4.1 Earthquake Source

With regard to the tectonic setting around Ambon, there are five candidate faults that could have generated a large earthquake with the intensities and effects shown in Fig. 2: (1) the north Seram Megathrust, (2) the Kawa Fault, (3) an intraslab fault (4) the purported South Seram Thrust Fault, or (5) a local fault on Ambon. Each of these is qualitatively analysed below to identify the most credible source of this event.

In general, the strongest ground motions were felt on Laitimor, Oma (Haruku), and Nusa Laut (Fig. 2). The intensity of ground motion decreases toward Banda Neira in the south, and Boano in the north. Many buildings collapsed and ground cracked in various places on Laitimor. There was liquefaction at Hutumuri and Waytome Rivers according to the accounts of ’water spurt[ing] high [in]to the air’. Recurring aftershocks were reported until at least 10 May. Judging by these observations, the source must have been a moderate to large earthquake with a local and shallow epicentre.

A shallow earthquake on the north Seram Megathrust would have been too far from Ambon to have these effects and if the earthquake had occurred on this fault, the islands of Boano, Kelang, and Manipa where moderate shaking was reported (Fig. 2) should have experienced stronger ground shaking than Ambon and Banda Neira. If the earthquake had originated from the Kawa Fault, the villages on Seram Island would have experienced more intense ground shaking as in the 1899 earthquake and tsunami event (Soloviev and Go 1974). A deep intraslab earthquake would generate ground motion felt over a broader region. However, deep intraslab earthquakes typically do not cause ground cracking and long aftershock sequences. The Benioff zone is over 100 km beneath Ambon (Spakman and Hall 2010), and an intraslab earthquake at this depth would likely generate strong ground motion distributed over a wider area than observed.

Therefore, credible sources for this event could have be the South Seram Thrust Fault or a local fault on Ambon. The South Seram Thrust was used in seismic and tsunami hazard maps of Indonesia (Irsyam et al. 2010; Horspool et al. 2014), where it is indicated as running from the south shore of Buru to Nusa Laut with northward dip (Fig. 1). Although there is apparently evidence for this fault in proprietary marine seismic survey results (J. Griffin, personal communication 2018), we note that it does not appear on the most recent revision of Indonesia’s seismic hazard map (Irsyam et al. 2019). We therefore regard the existence of this fault as speculative, even though an earthquake on it could have caused the observed ground shaking. Brouwer (1921) and Watkinson and Hall (2017) have confidently identified quaternary faults on Ambon itself, one of which Harris and Major (2017) identified as the source of the 1674 earthquake without a clear explanation of why. We conclude that any of these faults on and near Ambon could have generated the observed ground shaking, but further investigation is required to determine exactly which one ruptured in 1674.

4.2 Tsunami Source

If we consider which of the five candidate faults previously mentioned might have been capable of directly generating the observed tsunami, it is immediately obvious that an intraslab event could not, since an earthquake deeper than 100 km in the Benioff zone beneath the Banda islands could generate only a weak tsunami at best. Okal and Reymond (2003) showed that the largest intraslab event ever recorded, the 1938 \(\hbox {M}_w=8.5\) Banda Sea event at 60 km depth, generated only a weak tsunami. A tsunami generated by the Seram Megathrust or South Seram Thrust would have to enter Piru Bay through narrow straits around Ambon Island, which would greatly attenuate the tsunami arriving on the north coast of Ambon. As we show in Sect. 6 below, instead of having highest run-up along the north coast of Ambon, the tsunami energy generated by these two scenarios would be concentrated between the western and southern coast of Hitu and Laitimor, respectively. The Kawa Fault has a strike-slip mechanism that might generate a significant local tsunami generation where it crosses the southern shore of Seram, such as occurred in the 1994 Mindoro (Imamura et al. 1995) and 2018 Palu (e.g. Jamelot et al. 2019) events. However, it seems unlikely this mechanism could generate high tsunami run-up only along the northern coast of Ambon.

The only fault that might generate a large tsunami on the northern coast of Ambon and nowhere else would be a local fault documented on the northern shore of Hitu (Brouwer 1921; Watkinson and Hall 2017). The fault has a normal mechanism that could generate vertical displacement of a water column for tsunami generation. However, the fault under consideration is only 16 km long, and therefore unlikely to generate an earthquake with magnitude < 7 and slip much higher than 2 m (Kanamori and Anderson 1975; Geller 1976; Wells and Coppersmith 1994), far too small to generate the observed tsunami run-up. Therefore, if none of the potential tsunamigenic earthquakes could have been capable of directly generating a tsunami with the observed run-up, the most plausible source of the tsunami is from an earthquake-triggered landslide in Piru Bay.

There is an indication in Rumphius’ accounts 1997 that a massive coastal landslide occurred on the northern shore of Hitu, between Seith and Hila.

The country around Lebalehu, a region once famed for its market and for being the most important Muslim meeting place, collapsed the width of a Musquet shot. There is no longer any beach there, but only a very steep precipice. Just the same is true between Ceyt (Seith) and Hila, even as far as the beach at the later place, along the west side of the Fort Amsterdam and beneath the Residence of Intche Tay. Including the Negeris Nukunali, Taela and Wawani, all this disappeared along with [the] roadstead where ships used to anchor. It seems likely that the aforementioned wall of water arose in the place just indicated, to wit directly below Lebalehu. It might even have come from Hitu because various people on board ships that were not far off shore, reported only a rippling of the waves. The mass of upwelling water divided into three parts. One went east to Ceyt (Seith) and Hila, the other west to the villages of Lima and Oried. The water stank so horribly that people on board ships close to the coast became ill, and it was so filthy that anyone who had been immersed in it looked as if he had been hauled out of a mudbath.

In other words, there was a major change in the landscape around the coastal region: a gentle beach became very steep. Moreover, people saw that the water was dark and a roaring sound was heard from this area, which indicates that the seawater mixed with sediment and the tsunami source was near the people. The seawater colour and loud sound described in the accounts are similar to those described after the 1998 Aitape, Papua New Guinea (PNG) tsunami (Davies et al. 2003), which is thought to have been generated by a landslide. Unfortunately, there is no further information with regard to parameters of the landslide. According to the accounts, it was a partially subaerial landslide, with about 200 m of the previous shoreline collapsing into the sea. The lateral extent is thought to be at least 5 km along the coastline between the villages of Seith and Hila.

Further, the majority of the fatalities in this event were in this region and not many buildings collapsed in the earthquake. In addition, an extreme run-up laterally along the coastline can only be explained by a landslide. These observations are similar to the 1998 Aitape, PNG tsunami, which was generated by an underwater landslide (Okal and Synolakis 2004; Synolakis et al. 2002).

5 Landslide-Generating Tsunami

Tsunami generation is affected by both the vertical and length extent of the displace bathymetry (Heidarzadeh et al. 2014). Vertical displacement due to an earthquake normally reaches only a few metres at most, whereas it can easily reach hundreds of metres in a mass failure event. The dimensions of an earthquake rupture can extend to hundreds of kilometres but it is rare to see a huge landslide up to 100 km length. Therefore, a tsunami generated by an earthquake has a long wavelength as it travels across the ocean. On the other hand, a tsunami generated by a landslide loses its energy quickly as it travels because of its shorter wavelength, but it has a much larger amplitude in a local area. Therefore, tsunamis generated by earthquakes can be observed over a wide area, with a relatively constant run-up along the coastline. In contrast, a narrow region of extreme run-up is observed in a tsunami event generated by a landslide.

Okal and Synolakis (2004) investigated tsunami height profiles along a coastline from some past tsunami events. They found similar patterns in the 1946 Unimak (Alaska) and 1998 Aitape (PNG) tsunami profiles, which were both generated by landslides. The profiles had extreme run-up heights over a very narrow along-shore extent. Based on these findings, Okal and Synolakis (2004) developed a criterion for identifying landslide-generated tsunami based on the aspect ratio \(I_2=\frac{b}{a}\) between the maximum tsunami run-up height b and the distance a of the lateral extent of high tsunami run-up along the coastline. Thus, any event with \(I_2\) larger than 0.0001 should be considered a tsunami generated by a landslide.

Accordingly, we calculate the \(I_2\) of the tsunami observations on the northern coast of Hitu. It was difficult to determine the exact values of a and b because of the sparse data available (Fig. 3a). However, the data clearly shows a very rapid drop in run-up to the east and west of the maximum run-up height location between Lima and Hila. The estimated \(I_2\) was > 0.006, which was 60 times larger than the limit suggested by Okal and Synolakis (2004). Therefore, this tsunami almost certainly was generated by a landslide.

Fig. 3
figure 3

Tsunami height profiles along the coastline: a black squares are data; the dashed lines indicate \(I_2\) of \(a= 3\)–5 km with \(b= 40\)–100 m; b only from tsunamigenic earthquake scenarios; c selected landslide-generating tsunami scenarios. Symbols with colour indicate the scenario codes as shown in Tables 3 and 4

Full size image

6 Tsunami Modelling

Tsunami modelling was performed to confirm the analysis above using the JAGURS tsunami simulation code (Baba et al. 2015, 2017). The code numerically solves the shallow water wave equations in a spherical coordinate system with a finite-difference scheme. The digital elevation model (DEM) was built from a combination of nautical charts, a 90-m commercial bathymetry dataset provided by the TCarta Marine, the General Bathymetric Chart of the Oceans (GEBCO), and the SRTM-90m in a domain of nested grids (Fig. 4). The coarsest and finest grid resolution of the domain is approximately 1500 and 167 m, respectively. A time-step of 0.5 s is set to satisfy the Currant stability condition.

Fig. 4
figure 4

Domain model of tsunami modelling. The blue boxes are the nested-grids domain for the tsunamigenic earthquake model and red box shows the domain model for landslide-generating tsunami scenarios. Beach balls and black starts represent scenarios on Table 3

Full size image

Tsunami simulations were conducted for tsunamigenic earthquake (Table 3) and landslide-generated tsunami scenarios in this area (Table 4). The initial sea surface elevation from the earthquake scenarios were assumed to be equal to the earthquake deformation calculated from the Okada (1985) formula, using the parameters discussed below.

A two-layer model was utilised to simulate tsunami generation and propagation due to a landslide (Baba et al. 2019). As with the single-layer, earthquake-generated tsunami simulation, the two-layer simulation requires specification of ocean depth (i.e., bathymetry). In addition, the two-layer simulation requires an initial condition for the depth to the top of a viscous fluid layer, which is allowed to flow in response to gravity, exciting a tsunami by displacing the ocean layer above it (see Baba et al. (2019) for further details of the two-layer algorithm). The two-layer simulation was performed in a single grid domain model, over a tsunami propagation time of 2 h with time-step of 0.25 s.

Table 3 Tsunamigenic earthquake scenarios for the 1674 Ambon event
Full size table

6.1 Tsunamigenic Earthquake Scenarios

SMT-1 and SMT-2 (Fig. 5) represent two earthquake scenarios from the Seram Megathrust zone. The \(\hbox {M}_w=8.2\) parameters of SMT-1 (Løvholt et al. 2012) were speculated to be the source of the events documented in 1657, 1708, 1876, and 1965 (note that here and in subsequent sections, we use \(\hbox {M}_w\) to denote an input parameter for tsunami modelling, not a magnitude estimated for a historical earthquake). According to Horspool et al. (2014) and Irsyam et al. (2010), the Seram Megathrust zone is capable of generating an earthquake with a maximum \(\hbox {M}_w= 7.9\) to 8.2. Here, SMT-2 represents an earthquake with \(\hbox {M}_w= 8.1\) with the fault parameters taken from Horspool et al. (2014).

SST-1 and SST-2 (Fig. 5) represent two scenarios from the South Seram Thrust Fault. Løvholt et al. (2012) assumed the 1674 Ambon tsunami was caused by an \(\hbox {M}_w= 8.1\) earthquake on the South Seram Thrust Fault. The authors suggested that the same fault zone triggered the 1950 and 1983 events. The source parameters of the SST-1 scenario were taken from their study. Another study conducted by Latief et al. (2016) indicated the source of the 1950 Ambon tsunami as being on the South Seram Thrust Fault, but it was located a little further south, with a smaller fault plane. The SST-2 scenario followed the parameters used in Latief et al. (2016) and is represented as an \(\hbox {M}_w= 7.5\) earthquake.

Finally, ANF-1 and ANF-2 (Fig. 5) were designed to follow the normal fault, as indicated by Brouwer (1921) and Watkinson and Hall (2017), respectively. These two scenarios represent earthquakes with \(\hbox {M}_w= 6.0\) to 6.5, respectively. They are only hypothetical scenarios because no detailed study has been conducted on these faults, other than categorisation as quaternary faults (Watkinson and Hall 2017). Here a dip of 70° was selected as the optimum angle of a normal fault earthquake to generate maximum vertical displacement.

Fig. 5
figure 5

Initial sea surface elevation from tsunamigenic earthquake scenarios shown in Table 3. SMT-1 and SMT-2 represent the Seram Megathrust; SST-1 and SST-2 are the South Serahm Thrust; and ANF-1 and ANF-2 are local faults on the Ambon scenarios. Each figure has a different colour scale

Full size image

6.2 Landslide-Generating Tsunami Scenarios

We tested 36 synthetic scenarios of landslide-generating tsunami models (Table 4). The landslide layer was assumed to have a Gaussian function on top of the recent DEM with a deformable material type. The radius and thickness of the Gaussian varied from 500 to 2500 m and between 100 and 300 m, respectively. It was located at three different places (A, B, C) near the shoreline between Seith and Hila. The locations were selected according to the accounts previously discussed. To satisfy the stability of the model, the landslide layer needed to be ‘clipped’, so that the landslide would always remain below the at-rest level of the sea surface. Therefore, the landslide layer looked like a Gaussian-shaped accumulation of sediment that is clipped where it emerges above the water surface (Fig. 6).

Table 4 Landslide-generating tsunami scenarios
Full size table
Fig. 6
figure 6

Illustration of selected tsunami landslide scenarios shown in Table 4. B10 and B12 are landslides with the centre of the Gaussian located at Point B (128.030° E and 3.583° S) with each radius of 2500 m and maximum thickness of 100 and 300 m, respectively. The middle and bottom figures are ’clipped’ initial bathymetry (green), final bathymetry (black), and the Gaussian (red) profiles along the W–E and N–S direction

Full size image

7 Results and Discussion

Piru Bay is a nearly closed sea surrounded by several narrow straits surrounding Ambon Island. It is difficult for much tsunami energy to propagate from outside the bay to Ambon’s northern shore, particularly in the SMT-1, SMT-2, SST-1, and SST-2 scenarios. Therefore, the maximum tsunami amplitude simulated was relatively small (Fig. 7). These results meant that a tsunami source inside the bay is required.

As suspected, no tsunamigenic earthquake model could generate a high tsunami inside Piru Bay only, with the maximum height located on the northern shore of Ambon, except the normal fault scenarios (Fig.  3b). However, the fault length limits the magnitude, requiring a huge slip to produce a very high tsunami. This would be an unrealistic model. The normal fault model scenarios in this study were already unrealistic, with a 5-m slip.

Fig. 7
figure 7

Maximum tsunami heights from the tsunamigenic earthquake scenarios shown in Fig. 5

Full size image

All the landslide-generating tsunami simulations result in a maximum tsunami height concentrated inside Piru Bay between Seith and Hila with minor tsunamis at the other places (Fig. 8). Through these simulations, we confirm the tsunami height distribution mentioned in the historical accounts (Rumphius 1675). The largest tsunami height near the shoreline is almost 80 m from the B12Footnote 1 landslide scenario, with an approximate volume of 1 \(\hbox {km}^3\) (Fig. 3c).

Fig. 8
figure 8

Selected maximum tsunami height from landslide scenarios shown in Table 4

Full size image

The 100 m tsunami run-up between Seith and Hila villages could not be reconstructed for several reasons. First, there was no high-resolution DEM available to accommodate detailed inundation modelling. Therefore, the tsunami height profiles were extracted along the depth of 20 m and Green’s law (Synolakis 1991) was used to estimate the run-up height on the shoreline. Second, any DEM would not represent the actual elevation at that time. Third, it was possible that the tsunami run-up reported had been exaggerated. For example, ’the water rose especially between those villages (Lima and Hila) and Seith to the top of the surrounding hills, estimated to be some 50 to 60 fathoms [90–110 m] high’. This account could have meant a maximum run-up height due to water splash. According to the recent DEM, the closest hill with a height above 100 m was located about 500 to 1500 m from the shoreline, meaning the northern coast had a narrow backed strip coastal by steep topography. Moreover, some accounts noted seawater rising as high as the window of a redoubt in Seith and over-topping Fort Amsterdam in Hila. According to the online photos available, all the redoubts and forts are located near the coastline and are two to three floors high (approximately 20–30 m). Therefore, our simulated landslide scenario produced a reasonable tsunami height.

Although the primary source of the ground motion was still unclear, it was most likely from a local and shallow earthquake. From the reports of ground cracking and building damage in Laitimor region, the earthquake location was most likely south of Ambon or Nusa Laut, with the northern limit in the Laitimor region. Further investigation is highly recommended.

7.1 Limitations

Because our DEM cannot resolve evidence of a slump, confirmation of this finding through a high-resolution bathymetric survey, as well as a more sophisticated landslide-generating tsunami model is needed. The two-layer model in the JAGURS code considered only a submarine landslide type. Therefore, as noted earlier, the landslide layer had to be ‘clipped’ to keep it below sea level, for code stability, which may have resulted in underestimating the actual condition. In addition, the code was a friction-less model, so the slide layer did not stop moving during the simulation.

7.2 Implications for the Other Historical Accounts

The primary source of this earthquake and tsunami was confirmed from the sparse historical accounts (Rumphius 1997). While the precise mechanism of the earthquake remains unclear, the source of the tsunami could be confirmed to be a coastal landslide.

This study has shown a technique for optimising sparse and incomplete accounts. The historical accounts of tsunamis in Indonesia have been documented in several catalogues (e.g. Wichmann 1918, 1922; Soloviev and Go 1974; Soloviev et al. 1986). Generally, all accounts of tsunami events began by describing ground motion felt at various places with different intensities. These were followed by tsunami heights observation along the coastline. For the 1674 Ambon event, there is no detailed information regarding the source of the ground motion, nor the tsunami. By using similar technique that used in this study, the primary source of the tsunamis noted in the catalogue might be changed and/or updated in more detail. For example, the primary source of the devastating 1899 Seram tsunami was associated with an earthquake from the Kawa Fault. The kind of analysis we have employed here might help resolve whether the tsunami was generated by coseismic displacement where the strike-slip fault crosses the shore, such as occurred in the 1994 Mindoro (Imamura et al. 1995) and the 2018 Palu (Jamelot et al. 2019) events, or was mainly caused by the landslides observed along the southern coast of Seram (Soloviev and Go 1974).

Investigating other historical accounts will help to reveal more about the most likely primary source of the tsunami in each region. This will allow a more comprehensive tsunami hazard assessment.

8 Conclusion

In this study we have shown how historical accounts of the 1674 Ambon tsunami, one of Indonesia’s largest and deadliest tsunami disasters, can be used to better understand its earthquake source and mechanism of generation. We have shown that, although the reports of ground motion intensity and damage fail to definitively identify the earthquake source, it is almost certain that it was local and shallow, probably either a crustal fault on Ambon itself or the putative South Seram Thrust Fault off its southern coast.

More significantly, we have shown that the only way to explain the extreme run-up only on the northern shore of Ambon, and in particular the very narrow lateral extent along the coast over which it occurred, is by attributing the tsunami generation to a submarine landslide. Tsunami scenario simulations showed that plausible earthquake sources could not generate such a run-up profile, but a submarine landslide of about 1 \(\hbox {km}^3\) volume, consistent with eyewitness accounts of dramatic changes in the coastal landscape, could produce run-up commensurate with the historical observations.

Our analysis of the 1674 Ambon tsunami suggests that, as suggested by more recent events like the 1992 Flores and 2018 Palu and Sunda Straight tsunamis (Pranantyo and Cummins 2019; Sassa and Takagawa 2019; Giachetti et al. 2012; Patton et al. 2018, respectively), submarine landslides are an important component of the tsunami threat in Indonesia and should be considered in future tsunami hazard assessments. We believe further work on this and other historical events, especially when combined with paleotsunami and bathymetric surveys, can provide important constraints on the unique nature of the tsunami threat in regions of particularly complex and active tectonics like eastern Indonesia.