Tsunami Occurrence 1900–2020: A Global Review, with Examples from Indonesia

We present an overview of tsunami occurrences based on an analysis of a global database of tsunamis for the period 1900–2020. We evaluate the geographic and statistical distribution of various tsunami source mechanisms, high-fatality tsunamis, maximum water heights (MWHs) of tsunamis, and possible biases in the observation and recording of tsunami events. We enhance a global statistical overview with case studies from Indonesia, where tsunamis are generated from a diverse range of sources, including subduction zones, crustal faults, landslides, and volcanic islands. While 80% of global recorded tsunamis during 1900–2020 have been attributed to earthquake sources, the median MWH of earthquake tsunamis is just 0.4 m. In contrast, the median water height of landslide tsunamis is 4 m. Landslides have caused or contributed to 24% of fatal tsunamis. During 1900–2020, more tsunamis with water heights > 1 m occurred in Indonesia than in any other country. In this region fatal tsunamis are caused by subduction zone earthquakes, landslides, volcanos, and intraplate crustal earthquakes. Landslide and volcano tsunami sources, as well as coastal landforms such as narrow embayments have caused high local maximum water heights and numerous fatalities in Indonesia. Tsunami hazards are increased in this region due to the densely populated and extensive coastal zones, as well as sea level rise from polar ice melt and local subsidence. Interrelated and often extreme natural hazards in this region present both an opportunity and a need to better understand a broader range of tsunami processes.


Introduction
Tsunami research accelerated in response to devastating tsunamis that occurred in 1992 in Nicaragua and Indonesia. In the following three decades, additional tsunamis occurred that challenged the conventional assumptions for tsunami wave generation, propagation, and inundation (Kânoglu et al., 2020;Okal, 2019). The preponderance of shallow subduction zone faulting and generation of tsunamis naturally led to an early focus on these sources. However, tsunami research has expanded to include: (1) tsunami source mechanisms other than typical subduction zone earthquakes, including crustal strike-slip and thrust fault earthquakes, subaerial and submarine landslides, volcanic eruptions and meteorological events; (2) environmental effects on tsunami wave propagation and inundation (e.g. coastal and shelf geometry); and (3) climatic effects, such as sea level rise. The expansion of tsunami research has shown that tsunami generation and propagation are more complex than previously thought. For comprehensive hazard mitigation, particularly in a time of rapid climate change, it is important that all tsunami generation and propagation mechanisms are observed and understood.
It is now recognized that landslides in particular are highly effective at generating local tsunamis (Bardet et al., 2003). In the past few decades, the contribution of submarine and subaerial landslides to tsunami wave generation has been documented in cases where local runup observations are too high to be explained by fault displacement models alone (Harbitz et al., 2014). Methods of modeling tsunamis have also expanded to describe wave generation and propagation from different source mechanisms, including landslides (Løvholt et al., 2015;Okal & Synolakis, 2004). Furthermore, increasingly sophisticated numerical modeling has been employed to study the amplification effects of shelf and coastal geometries (Grilli et al., 2012).
Indonesia and the surrounding region ( Fig. 1) present an opportunity to study interrelated tsunami sources and effects, particularly those that are unaccounted for in current tsunami hazard mitigation strategies around the world. Tsunamigenic sources in Indonesia are diverse, ranging from earthquakes to volcanos to landslides, often in combination (Titov, 2021). Tsunamis in this region are also extremely hazardous. In the past 120 years, seven tsunamis in Indonesia, the Philippines, and Papua New Guinea caused over 1000 fatalities each (1907,1976,1979,1992,1998,2004,2018) (NCEI/WDS, 2021). Four of those events involved tsunamigenic landslides (1979,1992,1998,2018) (NCEI/WDS, 2021). The tsunami record of the Indonesia region demonstrates the need for assessment of local tsunami hazards, especially subaerial and submarine landslides, wave amplification effects of nearby bathymetry and landforms, and local relative sea level rise. We investigate these hazards by analyzing a compilation of global tsunami sources and effects, including maximum runups and fatalities, between 1900 and 2020. We include topographic, bathymetric and population data to study the interrelated factors of high tsunami hazard in Indonesia. Fatalities and high runups associated with various tsunami sources in Indonesia are compared to global data. We review tsunami occurrence in a diverse range of tectonic environments within Indonesia and discuss the tsunamigenic role of past and potential landslides. This work focuses on the three most common causes of tsunamis: earthquakes, volcanoes, and landslides, all of which are geologic processes. Rarer but also hazardous ''meteotsunamis'' are waves of comparable wavelengths to geologic tsunamis but are generated by atmospheric pressure changes. For a review of meteotsunamis, see Rabinovich (2020).

Data and Methods
Tsunami data recorded during 1900-2020 were obtained from the Global Historical Tsunami database maintained by the National Oceanic and Atmospheric Administration (NOAA) National Centers for Environmental Information/World Data Service (NCEI/WDS), referenced hereafter as NCEI/ WDS (2021). This database contains two linked catalogs: (1) recorded tsunami events from 2100 BCE to the present, and (2) runup observations corresponding to those tsunami events. Tsunami observations including runup and water height are recorded where observed in the NCEI/WDS database. ''Water height'' refers to the maximum elevation in meters of the wave above a reference sea level, whereas ''runup'' refers to the maximum elevation in meters of ground wetted by the tsunami (NCEI/WDS, 2021). These observations are often but not always recorded at the furthest inundation point inland. While in some cases we refer specifically to runup observations, we follow the example of the NCEI/WDS database in calling the maximum observed value from a given tsunami the ''maximum water height,'' (MWH) regardless of observation type.
An important consideration is the year when the tsunami occurred. Pre-1900 entries in the NCEI/WDS database are not comprehensive due to minimal instrumentation and inconsistent documentation of tsunami occurrence across the globe (Gusiakov, 2009). For this reason, in our statistical analyses we limit the dataset to events that occurred during 1900-2020 in order to take advantage of improved water height and earthquake monitoring technology, increased and standardized post-tsunami survey observations, and the globalization of record-keeping (Arcos et al., 2019;Synolakis & Okal, 2005). This limitation reduces the number of tsunamis we consider from 2761 to 1380 (as of December 2021; this database is dynamic and entries for past tsunamis are frequently added). While we believe that a restricted temporal range is a useful tool to reduce statistical bias, we recognize that it fails to provide the full historical context of tsunami occurrence in any given region.
Population density data for Indonesia, Papua New Guinea, and the Philippines were obtained from worldpop.org. This database estimates the number of people per grid unit for 2020, adjusted to match the UN national estimates (http://esa.un.org/wpp/). Location data of active Holocene volcanoes are from the Global Volcanism Program of the Smithsonian Institution (Global Volcanism Program, 2013). All maps were generated in QGIS (QGIS, 2021).

Global Perspective of Fatal and High Maximum
Water Height Tsunamis

Results of Global Analysis
Globally, large tsunamis occur relatively infrequently, with just 300 probable and definite tsunamis with an MWH [ 1 m recorded between 1900 and 2020 (NCEI/WDS, 2021). In those 120 years, the average frequency of recorded tsunamis [ 1 m is * 2.5/year. For the 20 year time period 2001-2020, an average of 3.4 tsunamis [ 1 m were recorded each year. The interquartile range (IQR) of tsunami MWHs is 0.16-3 m. However, rare and hazardous MWHs in excess of 20 m occur more frequently in climatically or tectonically extreme regions. This includes active hotspot volcanos, modern convergent zones, and high latitudes ([ 35°) ( Fig. 2) in fjords, such as those in Alaska, Greenland, and Norway. Between 1900 and 2020, twenty-two tsunamis with a MWH over 20 m occurred at latitudes [ 35°, compared to 5 events at latitudes \ 35°.
Regions with a high concentration of tsunami occurrence are demarcated in Fig. 3a and compared in Fig. 3b. Our analysis shows that globally, 80% of tsunamis are attributed to earthquakes (Fig. 3b). 12% are caused by landslides, 5% by volcanos, 2% by meteorological events, and 1% by unknown causes. While earthquake tsunamis are more commonly observed and generally better understood, landslide and volcano tsunamis have had disproportionately large MWHs (Table 1). The median MWH observed is 3.6 m higher for landslide tsunamis than for earthquake tsunamis, whereas the median MWH for volcano tsunamis is 2.1 m higher than the median for earthquake tsunamis. Landslides cause or contribute to 23% of all tsunamis with an MWH [ 1 m.
We observe that in general, fatalities are caused by tsunamis with MWHs [ 1 m, and fatalities generally increase with MWH ( Fig. 4). High-fatality tsunamis are geographically segregated by cause (Fig. 5a). Fatal landslide tsunamis occur at active and passive margins, frequently at high latitudes, while high-fatality earthquake tsunamis occur in active  disproportionately fatal tsunamis: landslides cause or contribute to 24% of the world's fatal tsunamis (Fig. 5b).

Discussion of Global Observations
Our findings from Table 1 are consistent with the previous analysis of tsunami databases by Gusiakov (2020), who found that non-seismic sources (i.e., landslide, meteorological or volcanic events) caused 36.7% of tsunamis with annual maximum water heights during 1900-2020. Furthermore, despite the fact that landslide-generated tsunamis and the resulting high runups are most commonly observed in sparsely populated high-latitude regions, we confirm that landslides are particularly hazardous, causing or contributing to nearly a quarter of fatal tsunamis. Moreover, while extreme runups are often observed in high latitude fjords where the dominant geomorphic force is glacial, extreme runups may also be Table 1 Distributions of maximum water heights (m) for tsunamis of different sources (1900-2020) Data are from NCEI/WDS (2021). The above distribution includes probable and definite tsunamis (values 3 and 4) recorded globally for which a maximum water height (MWH) was recorded. The total event count of 761 is lower than the total number of probable and definite tsunamis 1900-2020 (888) because 127 such events do not have recorded water height observations in the database (92 earthquake tsunamis, 23 landslide tsunamis, 8 volcano tsunamis, 1 meteotsunami, 3 tsunamis of unknown cause). ''Earthquake'' includes only pure earthquake sources (cause code 1), ''Landslide'' includes landslide and earthquake/landslide causes (cause codes 3 and 8), ''Volcano'' includes all tsunamis with some volcanic element (cause codes 4, 5, 6, and 7)  . a Global map of tsunamis by cause and number of deaths. Note that many fatal landslide tsunamis occur in the higher latitudes ([ * 35°N). Fatal earthquake tsunamis are prominent around the Pacific Ocean, as well as the northeastern Indian Ocean. Events with either a quantitative death count or a general description of nonzero fatalities are shown (127 events in total). b Percentage of fatal tsunamis by source mechanism. 24% of fatal tsunamis were caused at least partially by landslides.
Tsunami data are from NCEI/WDS (2021) Vol. 180, (2023) Tsunami Occurrence 1900-2020: A Global Review, with Examples from Indonesia 1555 expected in other regions where non-glacial processes have produced coastlines with similarly pronounced topography and morphology. An important difference between earthquakes and landslides lies in the facility of observation. A submarine landslide, for example, will go unnoticed unless a significant tsunami results (and the resulting waves diverge significantly from expectations based on an earthquake source model) and an investigation follows (e.g., Tappin et al., 2001;von Huene et al., 2014). This laborious and expensive forensic practice stands in contrast to earthquake detection, which is automated. This is a possible bias in comparing the frequency of landslide versus earthquake sources; landslide tsunamis \ 1 m are likely underdocumented.
The coastal impact from tsunamis is fundamentally linked to climate and sea level. Thick sedimentation, unstable pore pressure, and steep subaerial and submarine morphologies have been linked to increased tsunamigenic landslides at higher latitudes, where quaternary glaciation controlled coastal processes (Twichell et al., 2009). Sea level rise also has consequences for tsunami hazard. Equatorial and tropical regions are projected to undergo more extreme storm weather and flooding in the coming decades as a result of sea level rise (Vitousek et al., 2017). These and other effects of global warming and sea level rise may affect the geographic distribution of large and deadly tsunami events and cause particular damage in low latitude regions of complex-source tsunamis.
Between 1900 and 2020, 149 tsunamis were recorded in Indonesia, Papua New Guinea, and the Philippines (NCEI/WDS, 2021). Of those events, 85% have been attributed solely to earthquakes (Fig. 3b). This result is slightly higher than the global average (80%) but is consistent with the large number of seismogenic settings in this region.
In addition to earthquake sources, since 1900 ten tsunamis partially or fully caused by volcanoes occurred in this region, accounting for 7%, which is greater than the global average of 5% (Fig. 3b). Volcanic islands are widespread here (Fig. 1) and particularly tsunamigenic due to the proximity of water. A number of these islands are densely populated (Fig. 6). Furthermore, island arcs form chains of closely arranged and vulnerable coastlines.
While only 8% of recorded tsunamis during 1900-2020 were attributed to landslides in Indonesia, Papua New Guinea, and the Philippines, many of the highest recorded water heights, and some of the highest fatality tsunamis are associated with subaerial and submarine landslides. Tsunamigenic events in this region are distributed along subduction zones, particularly the Sunda and Java trenches, but also along transform and back arc faults such as the Palu-Koro fault and the Flores thrust, respectively (Fig. 7). During 1900-2020, seven tsunamis in the Indonesian region caused over 1,000 fatalities each (Fig. 7). Of these seven, three were caused by great (M w C 8.0) earthquakes: the 1907 Sunda trench earthquake and tsunami, the 1976 M w 8.0 Moro Gulf earthquake and tsunami, and the 2004 M w 9.1 Indian Ocean earthquake and tsunami (NCEI/WDS, 2021). The other four were caused fully or partially by landslides.
Earthquakes, landslides and volcanos in Indonesia and surrounding regions have produced tsunamis with runups in excess of 10-15 m (Fig. 8). In fact, between 1900 and 2020, more tsunamis with an MWH [ 1 m occurred in Indonesia than in any other country, including Japan, which records tsunamis most frequently (Fig. 9). The high number of tsunamis observed in Japan likely is due to extensive instrumentation, and results in a biased global distribution that does not reflect global comparative tsunami hazard. Furthermore, eight out of the twelve low-latitude (358 S-358 N) tsunamis with an MWH C 15 m since 1900 have occurred in Indonesia. Examples of these events include: (1)

Indonesian Tsunamis by Tectonic Setting
The following sections describe the tsunami history and present-day hazards in several regions of Indonesia. These regions serve as examples of different tectonic and geomorphologic environments, and the particular tsunami hazards associated with them. Figure 8 shows the locations of the five representative regions described below.

Subduction Zones
The Sunda megathrust is the dominant tectonic boundary of Indonesia in terms of length (5500 km; Fig. 9; Sieh, 2007). The Indo-Australian plate subducts beneath the Eurasian plate at a rate of * 68 mm/year near Java (DeMets et al., 2010), building a volcanic island arc that forms much of Indonesia's land-mass, including the islands of Sumatra and Java (Fig. 8). Like many subduction systems worldwide, the Sunda megathrust has generated numerous tsunamigenic earthquakes: in the past 120 years, 12 earthquakes along this boundary caused wave heights greater than 1 m. Six of these caused great loss of life ( Fig. 10; Table 2). Many of these events were shallow, slow rupture ''tsunami earthquakes'' (Kanamori, 1972), which have high  (2018) Vol. 180, (2023) Tsunami Occurrence 1900-2020: A Global Review, with Examples from Indonesia displacement and comparatively little ground shaking. In these cases, fault displacement is greater than 5 m and occurs at shallow subduction zone depths, which maximizes the displacement of the overlying water column and causes a large tsunami wave (Kanamori, 1972;Satake & Tanioka, 1999). However, fault displacement alone cannot account for extreme water heights observed for some of the Sunda megathrust tsunamis. For example, a tsunami in 2006 (Fig. 10) generated locally concentrated water heights of 10-21 m at Nusa Kambangan on the south coast of Java (Fritz et al., 2007;Lavigne et al., 2007;Mori et al., 2007). The great water heights and geographic localization are inconsistent with a tsunami caused by fault displacement and inferred to signify a local landslide contribution (Fritz et al., 2007;Lavigne et al., 2007;Mori et al., 2007). This inference of a probable landslide contribution near Nusa Kambangan to the 2006 Java tsunami is in keeping with the increased recognition since 1998 of the tsunamigenic role of coseismic landslides (Løvholt et al., 2015). Hébert et al. (2012) present a combined earthquake-landslide source that closely Evidence of past submarine landslides is incomplete and requires expensive marine surveys. The available evidence indicates that the number of submarine landslides along the Sunda megathrust increases to the southeast (Brune et al., 2010). It has been proposed that increased submarine landslides in this part of the forearc is linked to the regional characteristics of convergence (Brune et al., 2010). Convergence is perpendicular to the trench in the eastern portion of the Sunda megathrust (south of Central and East Java). Here, the forearc slope is steep (4.0°), and the subduction regime has been characterized as erosive (Kopp et al., 2006). Although evidence is still limited, preliminary studies suggest that erosive forearcs could be more susceptible to submarine landslides than accretionary forearcs (Kawamura et al., 2014). In contrast, at the NW Sunda trench (southwest of Sumatra, Fig. 10), convergence is oblique, and accretion of crustal material is extensive, resulting in a shallow 1.4°f orearc slope (Clift & Vannuchi, 2004). The Sunda megathrust and other subduction interfaces in Indonesia, Papua New Guinea, and the Philippines present valuable opportunities to explore this hypothesis. If a correlation exists, specific hazard mitigation strategies could be better tailored to different areas along subduction zones according to the risk of tsunamigenic landslides.

Continental Shelf
The bathymetry of the Indonesian region varies greatly, ranging from deep trenches at subduction zones to shallow seas. The interior of the Sunda shelf represents the latter, as it lies mostly within the Sunda block (Fig. 7) and contains several land masses including the southeastern Eurasian continent, western Sulawesi, Borneo, Sumatra and Java (Fig. 8). The Sunda shelf is characterized by shallow bathymetry that was mostly above sea level during the last glacial  (Hanebuth et al., 2000). Sediment-laden passive continental margins have been known to yield large, low angle landslides and debris flows, some of which have been tsunamigenic (Masson et al., 2010). A prominent example of this environment is the Makassar Strait (Fig. 11) situated between Borneo and the Sunda shelf to the west and Sulawesi to the east. The Mahakam delta deposits sediment into the western margin of the strait, while the eastern (Sulawesi) margin is characterized by a seismogenic fold-thrust belt (Prasetya et al., 2001). The Makassar Strait throughflow, a strong thermohaline current, flows from north to south. This current causes slopeparallel erosion and sedimentation along the Makassar Strait, oversteepening the deltaic deposits east of Borneo. This process of slope-parallel deposition is suspected to have caused at least 19 submarine paleolandslide deposits discovered on the eastern side of the strait (Brackenridge et al., 2020). These landslides may have generated large tsunamis. The nearby Brunei slide (Gee et al., 2007) in northwest Borneo is an example of a catastrophic submarine landslide that occurred in an environment of active rapid sedimentation, not unlike the Mahakam delta. The Brunei slide likely generated a catastrophic tsunami in the near field (Okal et al., 2011). The velocity of a submarine landslide is an important factor that affects tsunamigenic potential (Løvholt et al., 2015;Okal & Synolakis, 2004). In the   Satake et al., (2013) and Yue et al., (2014) Tsunami Fatalities are from NCEI/WDS (2021). Earthquake magnitudes and source descriptions are from references in the last column southern Makassar Strait, deposits related to a submarine landslide (Fig. 11) were studied by Armandita et al. (2015). They concluded from the cohesive quality of the deposit, despite the weak sediments involved, that this landslide failed slowly and did not generate a tsunami. However, it is not known whether a region with a history of slow submarine ground failure is capable of producing fast, tsunamigenic landslides. The steep slopes on both sides of the Makassar Strait, the regional history of landslides and seismicity, as well as the potential risk to coastal communities on either side of the strait, warrant further study of the probability of tsunamigenic landslides.

Figure 11
Landslide deposits in Makassar Strait. The red dotted line shows the location of 19 mass transport deposits identified by Brackenridge et al. (2020). The blue dotted line represents a mass transport deposit studied by Armandita et al. (2015). Note the sediment influx from the Mahakam Delta and over-steepened slopes on both sides of the strait, as well as tsunamigenic earthquakes on Sulawesi side. Earthquake symbols (stars) are scaled to the number of deaths recorded. Tsunami sources and water height observations are from NCEI/WDS (2021) 1562 J. A. Reid and W. D. Mooney Pure Appl. Geophys.

Crustal Faults
While subduction zone megathrust earthquakes are associated with most tsunamis in Indonesia (Fig. 8), crustal thrust and strike-slip earthquakes have also caused fatal, high-runup tsunamis in the recent past. Furthermore, the environments where these events occur can be conducive to amplification of propagating tsunami waves. Back-arc and strikeslip fault systems in this region accommodate strain by folding and incising oceanic and continental crust, which can lead to pronounced embayments and complicated landform geometries (Hall, 2012). These morphologies can be susceptible to landslides in addition to wave amplification. Additionally, arc volcanism in these regions provides steep slopes susceptible to failure, a tsunami triggering mechanism. Due to shorter faults and locally tsunamigenic mechanisms such as landslides and volcanos, a crustal fault tsunami may not have the long wavelength and impact radius of a large megathrust event. Nonetheless, as a result of the environmental factors outlined above, the local impact on coastal communities can be disproportionately severe.

Thrust Faults
In 1992, a M w 7.8 earthquake associated with the Flores back arc thrust (Fig. 7) struck Flores Island (Fig. 12). The rupture area has been proposed to lie roughly along the north coast of Flores Island (Pranantyo & Cummins, 2019). A devastating tsunami resulted, and together with the earthquake killed at least 2080 people (NCEI/WDS, 2021). Roughly half of these fatalities are attributed to the tsunami (Yeh et al., 1993). According to the fault model by Pranantyo and Cummins (2019) (Fig. 12), Tsunami Occurrence 1900-2020: A Global Review, with Examples from Indonesia the tsunami was generated very close to shore (0-30 km in most places), with little opportunity for the waves to attenuate. Water height observations were unevenly distributed alongshore and extremely high ([ 10 m) in some places (Fig. 12). Attempts to model the tsunami wave assuming that the source was solely due to the M w 7.8 earthquake alone have fallen short of observed water heights, particularly in reproducing runups of 26.2 m in Riang-Kroko, as well as 10-14 m in Hading Bay (Hidayat et al., 1995;Imamura et al., 1995;Pranantyo & Cummins, 2019). The conclusion is that coastal landslides likely contributed to these extraordinary water heights (Hidayat et al., 1995;Pranantyo & Cummins, 2019;Tsuji et al., 1995;Yeh et al., 1993). This event illustrates how back-arc thrusts and other near-shore Figure 13 Palu Bay tsunamigenic earthquake epicenters, tsunami water heights and geometry. While the 1927 and 1938 tsunamigenic earthquakes are shown, water height observations displayed are from the Sept. 28, 2018 tsunami (earthquake epicenter shown by yellow star). The highest waves were observed at the head of Palu Bay. Green outline shows where coastal landslides were observed, some of which were tsunamigenic. The primary Palu-Koro fault is shown, but the region is dominated by secondary faults, several of which were active during the 2018 earthquake (Sepúlveda et al., 2020). Tsunami data are from NCEI/WDS (2021) fault systems can both generate near-field tsunamis and trigger damaging tsunamigenic landslides. The 1992 Flores event highlights the vulnerability of island communities to tsunamis. Because earthquake tsunamis are caused by seafloor deformation, the resulting waves involve the entire water column, unlike wind-driven waves. Thus, tsunami waves interact differently with bathymetric features such as shoals and islands (Yeh et al., 1994). Following the 1992 earthquake, water heights of 5.5-7.1 m were observed on the southern coast of Babi Island ( Fig. 12; Tsuji et al., 1995), a part of the island that is normally protected from strong surface waves of the open ocean. These high water observations on the lee shore are attributed to positive interference from waves wrapping (refracting) around the circular island (Briggs et al., 1995;Kânoglu & Synolakis, 1998;Yeh et al., 1994). Small islands like Babi Island are common in the Indonesian region, many of which reside near active fault zones that could trigger tsunamis. Coastal communities that are protected from wind-driven waves and normally have quiet waters may be unaware of and vulnerable to the refraction properties of a tsunami wave. The 1992 Flores tsunami highlights the particular need for analysis of potential tsunami hazard for small islands across this region, including those located in the back arc.

Strike-Slip Faults
A recent example of a devastating tsunami event with a complex source is the 2018 M w 7.5 Sulawesi earthquake and landslide tsunami in Palu Bay, Sulawesi (Omira et al., 2019) (Fig. 13). The primarily strike-slip earthquake associated with the Palu-Koro transform fault caused intense ground shaking, liquefaction, onshore landslides and coastal failure (Cilia & Mooney, 2021;Omira et al., 2019). Due to their minimal vertical seafloor displacement, strikeslip earthquakes are not commonly the source of such devastating tsunamis (Ward, 1980). However, largely enclosed bodies of water such as Palu Bay (Sulawesi, Indonesia) are susceptible to significant wave generation from strike-slip fault motion, particularly a supershear earthquake such as the 2018 event (Elbanna et al., 2021). What's more, strike-slip faults near coastlines are capable of creating precisely such elongate bays, for example Tomales Bay (California, USA). In the case of Palu, coastal failures (Omira et al., 2019;Sassa & Takagawa, 2019;Sepúlveda et al., 2020) and submarine landslides (Carvajal et al., 2019;Pakoksung et al., 2019) have both been proposed as potential sources for the tsunami waves. The models that best fit observed water heights incorporate contributions from three sources, including (1) coastal landslides, (2) submarine landslides, and (3) submarine fault motion (Cilia et al., 2021;Sassa & Takagawa, 2019;Sepúlveda et al., 2020).

Normal Faults
Normal fault earthquakes have also been known to cause tsunamis, though infrequently. In Indonesia, the 1977 Sumba tsunami (Gusman et al., 2009;Fig. 10; Table 2) was caused by a shallow M w 8.3 normal-faulting earthquake in the subducting Indo-Australian plate just south of the Java trench (Lynnes & Lay, 1988). Outer rise earthquakes such as the 1977 Sumba earthquake occur under deeper waters than megathrust earthquakes. This increases the amplification effect of shoaling tsunami waves (Gusman et al., 2009), leading to high runups at shore.

Volcanic Sources
Arc volcanism in this region is widespread. Volcanos can rapidly produce material and deposit it at steep angles, causing unstable relief that is resolved by faulting and landslides (Moore et al., 1989). Volcanic activity can also be a seismic trigger for potentially tsunamigenic ground failure. A wellstudied example in Indonesia is Anak Krakatau (Fig. 14). In December 2018, after 6 months of elevated volcanic activity, the southwestern flank of Anak Krakatau collapsed (Borrero et al., 2020;Grilli et al., 2019;Ye et al., 2020). The failed side of the volcano slid rapidly into the water, causing a tsunami that radiated to nearby shores and caused extreme water heights (up to 85 m) on the nearby Verlaten, Lang, and Rakata islands (Fig. 14). This event, which killed 437 people, was catastrophic but not surprising, because the volcano was known to be unstable and potentially tsunamigenic. In fact, a modeled collapse of the southwestern flank and resulting tsunami were produced six years prior (Giachetti et al., 2012), with results that were closely rendered by the actual 2018 event. Before the 2018 tsunami, this volcanic island had erupted several times in recorded history, causing tsunamis and countless fatalities. The 1883 eruption killed 36,417 people, mostly from tsunamis that struck nearby Java and Sumatra (NCEI/WDS, 2021; Paris et al., 2014). Another example of a tsunamigenic volcanic island in this region is Palu'e Island, which caused a tsunami in 1929 (Fig. 11). Volcanic islands are commonly densely populated, or close to populated coastlines, thus local tsunami risk to nearby communities is high.  1970(Oppenheimer et al., 2019. Increased hazards due to sea level rise are a global concern, particularly for coastal and island communities, including those in Indonesia, Papua New Guinea, and the Philippines (Nicholls & Cazenave, 2010). Small islands and low elevation coastal zones in these regions are also densely populated (Worldpop, 2018). A compounding issue is the significant tectonic and anthropogenic subsidence that is taking place in some densely populated parts of Indonesia (Chaussard et al., 2013;Kuehn et al., 2009) and the Philippines (Rodolfo & Siringan, 2006), which increases the rate of local relative sea level rise.

Relative Sea Level Rise and Tsunami Hazard
Significant effects of sea level rise include increased flood risk (Vitousek et al., 2017) from storms and tsunami inundation, as well as permanent submersion of populated areas (Oppenheimer et al., 2019). The tropics in particular are projected to undergo increased frequency in extreme flooding events, due to as little as 5-10 cm of sea level rise (Vitousek et al., 2017). Sea level rise is a growing topic of interest for tsunami hazard projections in vulnerable regions, particularly on the timescale of community development planning (Meilianda et al., 2019). A new variation of probabilistic tsunami hazard assessment works to account for the influence of the rising sea level (Sepúlveda et al., 2021). In southern China, a projected 50 cm of sea level rise by 2060 is predicted to double the tsunami hazard (Li et al., 2018). Indonesia and surrounding regions face the challenges of low elevation coastal populations, high proportions of reclaimed land, and nearby tsunamigenic sources.
Rising sea level will not only affect tsunami inundation and population vulnerability; sea level rise will also have broader effects on geomorphologic and tectonic processes (Leatherman et al., 2000;Limber et al., 2018;McGuire, 2010). As sea level rise alters the coastal and shelf environment, local tsunami hazards may develop, including shallow bathymetry and irregular geometries that can cause wave amplification. Furthermore, sea level rise has implications for erosion of coastal cliffs (Limber et al., 2018) and offshore slope stability. Sea level rise may influence the frequency of submarine landslides by increasing failure preconditioning mechanisms such as storm wave loading and sedimentation patterns (Tappin, 2010). The role of climate-cycling sedimentation patterns in the occurrence of tsunamigenic paleolandslides is studied (Bondevik et al., 2019). However, concrete implications of climate change for future tsunami hazard remain a pertinent and largely unexplored research question.

Conclusions
An analysis of tsunami occurrence worldwide for the period 1900-2020 provides a broad overview of the distribution and maximum wave heights of tsunamis globally. We have documented the importance of landslide-sourced tsunamis, which pose an underestimated threat to many coastal communities. Our analysis leads to the following conclusions: (1) Fatalities can generally result from tsunamis with maximum water heights [ 1 m (Fig. 4). Indonesia has the greatest number of tsunamis [ 1 m (Fig. 9), making tsunami hazard mitigation a high priority for this country.
(2) In many regions deadly tsunamis have multiple causes (Figs. 3 and 5); Indonesia, the Philippines, and Papua New Guinea have experienced devastating tsunamis not just from megathrust earthquakes, but also from intraplate crustal earthquakes, landslides and volcanic eruptions. These sources can also occur in combination, such as a landslide triggered by an earthquake or volcano eruption. In this region, four out of the seven deadliest ([ 1000 deaths) tsunamis involved landslides. Three of these seven events were not in active subduction zones, where most tsunami research is dedicated (Fig. 7). (3) Whereas earthquakes alone cause the vast majority of all recorded tsunamis, landslide-generated tsunamis are disproportionately represented among those events that cause deaths. Landslides cause 24% of all fatal tsunamis (Fig. 5b). However, hazards from landslide tsunamis are underestimated due to the logistical difficulty to document submarine and subaerial landslides. A major contributor to the high fatalities is the great water height achieved by landslide tsunamis. The median maximum water height from landslide tsunamis is ten times greater than that from earthquake tsunamis (Table 1). Every tsunami with a maximum water height over 56 m was due in part to a landslide (Fig. 4). Vol. 180, (2023) Tsunami Occurrence 1900-2020: A Global Review, with Examples from Indonesia 1567

Funding
This project was funded by the Bureau of Humanitarian Affairs within the US Agency for International Development and the US Geological Survey, Earthquake Science Center.

Software availability
Map figures were made using version 3.16 of the open-source Geographic Information Systems software QGIS (https://www.qgis.org/en/site/).

Declarations
Conflict of interest The authors have no relevant financial or non-financial interests to disclose.
Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if changes were made. The images or other third party material in this article are included in the article's Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article's Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http://creativecommons.org/licenses/by/4.0/.
Publisher's Note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.