A series of surveys was conducted at the five affected areas explained in the previous section from 24 December 2018 to 3 January 2019. The 10 d survey was initiated to measure the impacts of the tsunami on Banten area. In this area, the team spent 6 d measuring tsunami flow depths, tsunami boulders, and damage to housing or buildings in the affected area. Another 4 d of the survey were spent in Lampung measuring the same data. Measurement of flow depths was done using a levelling staff and water pass to measure water marks, broken twigs, or stranded tsunami debris. A handheld GPS was used to locate the coordinates of the measured flow depths. The GPS was also used to measure limit of tsunami inundation. Some previous post-tsunami impact surveys also applied a similar method to this study, collecting flow depths, tsunami inundations, and building damage (see Fryer, 2011; Borrero et al., 2006; and Syamsidik et al., 2019a). A drone was also utilized in this survey to capture images from the tsunami-affected area. In total, we managed to measure 117 flow depths from both sides of the affected area. All data were stored in the Mendeley dataset (Syamsidik et al., 2019b). The survey was not performed at offshore islands around Mount Anak Krakatau, as the area was restricted by the government for any activity due to the threat of volcanic activity and the tsunamis.

The tsunami arrival at the coastal area was analysed based on water elevations measured at four tidal-gauge stations. The locations of the tidal-gauge stations can be seen in Fig. 1. To separate the tsunami wave data from the high-frequency data influenced by astronomical components, a low-pass filter was applied. Threshold frequency for the filtering was 0.0805 cycles per hour (cycle h-1), as suggested by Emery and Thomson (2001). Arrival times were interpreted based on the first peak of the wave recorded at the tidal-gauge stations. To confirm the conditions around the arrival times of the tsunami, we also performed a number of interviews in the local community. Three main questions were asked, i.e. regarding indications or some sign before the tsunami arrived, the number of waves, and the evacuation process. Ten people were interviewed on the Banten coasts and five in Lampung. Since the number of interviewees was limited, results of the interviews were meant to confirm the conditions before the tsunami arrival qualitatively. Furthermore, interviews were also conducted to identify locations of onshore boulders before the tsunami.

Velocity data were inferred from tsunami boulder transportation. Types of boulders were classified based on size and material composition. During the survey, they were dominated by two kinds of boulder materials, namely coral and rubble mound material from revetment structures. A description of the analytical solutions for inferring tsunami boulder transportation was given by Noormets et al. (2004). Tsunami boulders were measured for their dimensions, original locations (based on interviews and materials), and distance of transport. Tsunami boulder transportation velocities were inferred using Eq. (1), as suggested by Paris et al. (2010): 1umin⁡=2μmgCdAnρw, where umin⁡ is minimum estimated velocity (m s-1), μ is the friction coefficient, which is 0.7, as suggested by Noormets et al. (2004), g is gravitational acceleration (m s-2), Cd is the drag coefficient, which was considered to be 1.95, An is the areal of the boulder perpendicular to the tsunami flow direction, and ρw is water density (1027 kg m-3).

Distance travelled by the tsunami boulder depended on tsunami velocity, size of the boulder, and boulder material. The range of seawall material transported as a tsunami boulder was 3–4 m s-1 for sliding or rolling movement and 11–12 m s-1 if the boulder moved in the saltation mode (Nandasena et al., 2011; Paris et al., 2009).

An assessment of building damage was performed on 98 buildings; most of them were houses. Among the assessed buildings, 73 were confined masonry-brick-infill houses. This type of house used smooth bars as structural components and masonry fill for walls. The walls were strengthened by tie columns and tie beams at the tops. Most of walls had a thickness of 15 cm. The bricks were bound by about 2–3 cm mortars. Plain bars were used to connect the walls with windows and door elements. This type of house predominated in Banten and Lampung. A similar type of house was found to be the majority of houses in the 2004 Indian Ocean tsunami-affected area in Aceh. Around 70 % of the houses in Aceh were confined-masonry-brick-infill houses (Boen, 2005; Brzev, 2007). This study compared the tsunami fragility curve produced based on the 2004 Indian Ocean tsunami in Banda Aceh to a fragility curve composed based on the 2018 Sunda Strait tsunami, as the types of houses in both areas are similar.

The types of damage were classified into five damage states (DSs), as suggested by Suppasri et al. (2011) and Macabuag et al. (2016). The list of DS classifications can be seen in Table 1. To exclude houses altered due to clearing processes, the assessment was limited to the confined masonry-brick infill buildings (CM) and wooden houses. Types of houses in the Banten and Lampung affected areas were relatively similar. Most of the buildings located near the coastal areas functioned as residences, villas, or cottages.

Data collected were analysed using fragility functions to produce cumulative probabilities of damage caused by the tsunami. Equation (2) was used to calculate the fragility functions: 2P(x)=ϕx-μσ, where P is the cumulative probability of damage, ϕ is the standardized normal distribution function, x is the hydrodynamic feature analysed for damage (in this case, flow depth), and μ and σ are the mean and standard deviations of x, respectively.

The fragility function was developed to estimate the future impacts of a tsunami. In previous cases, the function was developed based on tectonic tsunamis (Koshimura et al., 2009; Suppasri et al., 2015). In this study, as there was no earthquake preceding the tsunami, analysis of the damage was based purely on tsunami wave propagation.

Tsunami arrivals were determined based on water elevations monitored at four tidal-gauge stations located around the Sunda Strait. The stations were Marina Jambu, Ciwandan, Panjang and Kota Agung. The tidal-gauge stations are operated by Indonesia Geospatial Information Agency (BIG, 2018). Each of the stations recorded 1 min interval water levels and are available online. Figure 2 shows the filtered tidal data from all four stations. The first detected tsunami wave arrival was recorded at Marina Jambu station at 14:30 UTC (at 21:30 local time). The time of generation of the tsunami from the source was not clear, since no tidal station was located around the volcano. Thirteen minutes later (14:43 UTC), the tsunami wave was recorded at the Ciwandan station, near Cilegon, Banten. The first tsunami wave was recorded at 14:40 UTC at the Kota Agung station, which was almost simultaneous with the Ciwandan station. Although Panjang is closer to the Mount Krakatau complex than Kota Agung, due to tsunami wave arrays that caused diffraction and refraction, the arrival time at Kota Agung was 10 min earlier than at Panjang stations (at 14:50 UTC). The landslide area at the southern part of Mount Anak Krakatau also contributed to arrival times.

Comparing the two main affected areas, a significant flow depth decay on both sides of the study areas was found. The highest flow depth was found at Cipenyu, Banten, where the tsunami wave hit directly from the area around the Krakatau complex. In the southern and the northern parts of this location, flow depth decreased. This was also found to be true in Lampung. Detailed findings from both of these areas are elucidated as follows.

Most buildings in the surveyed areas can be classified into two types, namely confined masonry and wood. Wall components were brick. Roofs were wooden-framed, with either tiled or zinc roofing. Detailed investigation of building damage was performed at 98 sites, including the remaining houses and one school. Seventy-three of them were non-engineered, lightly reinforced concrete houses or confined masonry (CM), and 25 were wooden timber. There were 16 CM-type houses as classified by DS0. Impacts of tsunami debris were minor at houses near the coastline. Some tsunami debris was found to have contributed to house damage, although most of the CM-type houses were mostly unaffected. Large debris was found at hotel complexes in Banten, where cars were carried by the tsunami flow and stopped by vegetation around the complexes. Damage to CM walls was mostly due to impulse or punching force produced by the tsunami flow. Unlike cases with tectonic events, shear cracks on walls due to lateral forces were not found. Damage to columns was observed as collateral failure of walls combined with impacts of hydrodynamic forces. Figure 13 shows completely destroyed houses (DS4), where their types and material composition were known; these were also directly within tsunami flow depths as measured around the houses. Due to the limited number of surveyed houses, the fragility function was analysed using CM-type houses. A summary of the fragility functions analysis is given in Table 2. Fragility curves for all four types of damage can be seen in Fig. 14.

Fragility curves in Fig. 14 reveal that the absolute damage due to the tsunami waves occurred at a flow depth of 6.5 m. Lower probabilities of complete damage were found in this study compared to the 2004 Indian Ocean tsunami in Banda Aceh (Koshimura et al., 2009). These findings were significantly different from the 2018 Palu tsunami (Paulik et al., 2019), where earthquake and liquefaction (liquefied gravity flows) also contributed to the damage (Sassa and Takagawa, 2019).

There were 25 wooden houses surveyed in the affected area. Due to the limited number of surveyed wooden houses, analysis of the damage was performed by classifying the damage as seen in Table 1. Only three classifications of damage could be identified, namely DS2, DS3 and DS4; 2 houses could be classified as DS2, 7 houses were DS3, and 19 houses were DS4. Examples of the damage found at wooden houses due to tsunami impacts can be seen in Fig. 13. Based on the number of surveyed wooden houses, it was found that a tsunami flow depth higher than 2.0 m could have completely washed away the houses, provided there was no debris in the flow. If the tsunami flow contained debris, the limit of the tsunami flow depth that could wash away a house was lower. In cases where the flow depths were lower than 2.0 m, damage was found to be due to failure of the bottom plates of the walls. Similar investigations of the impacts of tsunami bores on wooden walls were completed by Linton et al. (2013) and Wilson et al. (2009). They revealed that (1) the tsunami flow at the foundations beneath the walls could generate an uplifting force and (2) if the wave broke near the wooden structure, it also increased loading. Although the location of the tsunami breaking wave in the 2018 Mount Anak Krakatau tsunami remains unknown, the pattern of damage to the lower parts of the wooden walls was similar to that in these studies.

In Banten, tsunami flow depths were found to be higher than those measured in Lampung on the Sumatra side. Although Mount Anak Krakatau is nearly equidistant to the Lampung and Banten coasts, flow depths were apparently higher in Banten. Furthermore, impacts of the tsunami were more severe on the Banten coasts. As the event was caused by a non-tectonic process at night, a large number of people fell victim to the unavailability of tsunami warnings that could have anticipated a volcanogenic tsunami. Significant flow depth decay found in this study was similar to that found during the 2002 Stromboli tsunami, as these were typical landslide tsunamis (Tinti et al., 2006; Okal and Synolakis, 2003).

The volcanogenic tsunami is a discernible event. There have been hundreds of tsunami events recorded in Indonesia since the 16th century, with only 11 % generated by a volcanic eruption (Latief et al., 2000; Puspito and Gunawan, 2005; Løvholt et al., 2012; Pribadi et al., 2013). The 22 December 2018 Sunda Strait tsunami was an apparent unanticipated event caused by the southwest flank failure of Mount Anak Krakatau. There was no proper warning released by the authorities, as the present tsunami warning system cannot anticipate non-tectonic tsunamis, as in the case of the 2018 Sunda Strait tsunami. Furthermore, tsunami arrivals were presumably on short notice. As in the case of the 2002 Stromboli volcanogenic tsunami, arrival times were 6 min at a distance similar to that of the 2018 Mount Anak Krakatau tsunami (Bonaccorso et al., 2003). This problem was evident during the 28 September 2018 tsunami event (Syamsidik et al., 2019a; Takagi et al., 2019). Following the 2018 Mount Anak Krakatau tsunami, the government of Indonesia has been planning to install a number of tidal gauges to monitor water elevation around the volcano complex associated with tsunami wave generation. Nonetheless, it should be noted that, despite the installation, it will still be difficult to detect tsunami waves with this method without other supporting sensors. This was proven in the case of Mount Stromboli in Italy (Tinti et al., 2003).

To anticipate tsunamis on the Banten coast, there is one official escape building in Labuan. It is located about 420 m from the coastline. During the 22 December 2018 event, this building was used by the community as a shelter despite being far from the affected area. The port of Labuan was hit by a tsunami wave at 1.10 m. There was no major damage found at this location. The location of the tsunami escape building in Labuan needs to be reconsidered, and further steps are required to properly accommodate the tsunami evacuation process. Lessons learned from tsunami escape buildings in Aceh and coastal settlements after the 2004 Indian Ocean tsunami should be taken into account (see McCaughey et al., 2017; Syamsidik et al., 2017).

Damage to buildings and houses was more severe in Banten than in Lampung. Most of the damage was found in wall components. As the majority of buildings in Banten and Lampung were constructed with brick walls, the masonry-infill walls had a 50 % probability of damage when tsunami flow depth reached 2.3 m and would certainly have experienced major damage had the flow depth reached 5.05 m. Unlike damage found in the cases of tectonic tsunamis, diagonal cracks in the buildings resulting from lateral forces were not found. The CM-type house would be completely washed away by the tsunami when flow depth reached 6.6 m. These findings demonstrated a slightly different pattern compared to tectonic tsunamis (Koshimura et al., 2009; Paulik et al., 2019). It is worth noting that the latter two studies examined a combination of tectonic and tsunami forces. Damage could even start when the flow depth has yet to reach its maximum level (Suppasri et al., 2019). A lower flow depth that could cause complete damage to the CM-type house was shown by Shoji et al. (2014). However, this was a non-probabilistic study, with only 25 sample houses observed. In the case of the 2018 Mount Anak Krakatau tsunami, the wavelength could have been shorter, as this was similar to a landslide-driven tsunami. This would have generated a smaller impulsive force on buildings. Therefore, this could also be a further explanation for the difference in fragility curves derived from the 2004 Indian Ocean tsunami compared to the 2018 Mount Anak Krakatau tsunami.

No building was overturned by the tsunami wave as in the case the 2011 Tōhoku earthquake and tsunami (Yeh et al., 2013; Latcharote et al., 2017). This proved that the damage to the buildings was generally due to impact and drag forces. Debris-laden flow forces did not impact most of the damaged houses located close to the coastline. Hydrodynamic forces decreased as the speed of the wave decreased due to the roughness coefficient of the land use types (Saatçioğlu et al., 2006). A specific case was found at the Mutiara Carita Cottages (see Figs. 5 and 6). Here, the force became larger when gaps between buildings were smaller. This was also found in some cases in Banda Aceh during the 2004 Indian Ocean tsunami (Triatmadja and Benazir, 2014). Debris generated mostly by the fronts of buildings facing the tsunami wave caused severe damage to other buildings downstream of the tsunami flow.

Limitations of this study can be attributed to several aspects. The first limitation was that the time of the tsunami generation from the Mount Anak Krakatau was still unknown when this article was submitted. There was no tidal gauge or sensor placed around the volcano complex that could have detected the first generation of the tsunami wave near the source. The second limitation was that the volcanic mechanism that generated the tsunami is still not known. There are four mechanisms that could have potentially generated the tsunami, such as underwater explosion, flank failures (landslide), pyroclastic flow and volcano crest collapse. The most likely process that generated the 2018 Mount Anak Krakatau tsunami was the last one. This is based on aerial images captured before and after the tsunami. However, whether other mechanisms could also have contributed to the tsunami could not be verified at the time of publication. The third limitation was that estimation of the velocity was based on tsunami boulder transport, whereas the calculation was based on experimental formulae. This would result in an estimated tsunami velocity rather than a recorded velocity. Tsunami fragility curves were drawn based on a limited number of tsunami-affected houses. Additional explanation and figures from the surveys can be found in TDMRC (2018, 2019). Notwithstanding these limitations, this study conclusively revealed the impacts of a pure tsunami wave force on buildings and could be useful for further studies in tsunami engineering and mitigation.